first commit

.gitattributes

@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.bmp filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,3 @@
+*.pth
+*.zip
+outputs/*

CODE_OF_CONDUCT.md

@@ -0,0 +1,80 @@
+# Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to make participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, sex characteristics, gender identity and expression,
+level of experience, education, socio-economic status, nationality, personal
+appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies within all project spaces, and it also applies when
+an individual is representing the project or its community in public spaces.
+Examples of representing a project or community include using an official
+project e-mail address, posting via an official social media account, or acting
+as an appointed representative at an online or offline event. Representation of
+a project may be further defined and clarified by project maintainers.
+
+This Code of Conduct also applies outside the project spaces when there is a
+reasonable belief that an individual's behavior may have a negative impact on
+the project or its community.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at <opensource-conduct@fb.com>. All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see
+https://www.contributor-covenant.org/faq

CONTRIBUTING.md

@@ -0,0 +1,32 @@
+# Contributing to OVSeg
+We want to make contributing to this project as easy and transparent as
+possible.
+
+## Pull Requests
+We actively welcome your pull requests.
+
+1. Fork the repo and create your branch from `main`.
+2. If you've added code that should be tested, add tests.
+3. If you've changed APIs, update the documentation.
+4. Ensure the test suite passes.
+5. Make sure your code lints.
+6. If you haven't already, complete the Contributor License Agreement ("CLA").
+
+## Contributor License Agreement ("CLA")
+In order to accept your pull request, we need you to submit a CLA. You only need
+to do this once to work on any of Meta's open source projects.
+
+Complete your CLA here: <https://code.facebook.com/cla>
+
+## Issues
+We use GitHub issues to track public bugs. Please ensure your description is
+clear and has sufficient instructions to be able to reproduce the issue.
+
+Meta has a [bounty program](https://www.facebook.com/whitehat/) for the safe
+disclosure of security bugs. In those cases, please go through the process
+outlined on that page and do not file a public issue.
+
+
+## License
+By contributing to OVSeg, you agree that your contributions will be licensed
+under the LICENSE file in the root directory of this source tree.

GETTING_STARTED.md

@@ -0,0 +1,99 @@
+## Getting started with OVSeg
+
+
+### Try demo
+
+We release our largest model (Swin-Base + CLIP-ViT-L/14) [ovseg_swinbase_vitL14_ft_mpt.pth](https://drive.google.com/file/d/1cn-ohxgXDrDfkzC1QdO-fi8IjbjXmgKy/view?usp=sharing) (md5: <tt>526080</tt>).
+
+- Test on sample image
+  ```bash
+  python demo.py --config-file configs/ovseg_swinB_vitL_demo.yaml --class-names 'Oculus' 'Ukulele'  --input ./resources/demo_samples/sample_03.jpeg --output ./pred --opts MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth 
+  ```
+
+### Evaluation with pre-trained weights
+
+We release our largest model (Swin-Base + CLIP-ViT-L/14) [ovseg_swinbase_vitL14_ft_mpt.pth](https://drive.google.com/file/d/1cn-ohxgXDrDfkzC1QdO-fi8IjbjXmgKy/view?usp=sharing) (md5: <tt>526080</tt>).
+
+- Test on ADE20K-150 and ADE-847
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth DATASETS.TEST \(\"ade20k_sem_seg_val\",\"ade20k_full_sem_seg_val\"\) 
+  ```
+
+- Test on PascalContext-59 and PascalContext-459
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth MODEL.CLIP_ADAPTER.CLIP_ENSEMBLE_WEIGHT 0.6  DATASETS.TEST \(\"pascal_context_59_sem_seg_val\",\"pascal_context_459_sem_seg_val\",\)
+  ```
+
+- Test on PascalVOC-20
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth MODEL.CLIP_ADAPTER.CLIP_ENSEMBLE_WEIGHT 0.45  DATASETS.TEST \(\"pascalvoc20_sem_seg_val\",\)
+  ```
+  
+#### Performance benchmark
+
+| method                             | backbone | training dataset | A-847 | PC-459 | A-150 | PC-59 | PAS-20 |
+|------------------------------------|----------|------------------|:-----:|:------:|:-----:|:-----:|:------:|
+| Open-vocabulary generalist models. |          |                  |       |        |       |       |        |
+| SPNet                              | R-101    | PASCAL-15        |   -   |    -   |   -   |  24.3 |  18.3  |
+| ZS3Net                             | R-101    | PASCAL-15        |   -   |    -   |   -   |  19.4 |  38.3  |
+| LSeg                               | R-101    | PASCAL-15        |   -   |    -   |   -   |   -   |  47.4  |
+| LSeg+                              | R-101    | COCO Panoptic    |  2.5  |   5.2  |  13.0 |  36.0 |  59.0  |
+| SimBaseline                        | R-101c   | COCO-Stuff-156   |   -   |    -   |  15.3 |   -   |  74.5  |
+| ZegFormer                          | R-50     | COCO-Stuff-156   |   -   |    -   |  16.4 |   -   |  80.7  |
+| OpenSeg                            | R-101    | COCO Panoptic    |  4.0  |   6.5  |  15.3 |  36.9 |  60.0  |
+| OVSeg (Ours)                       | R-101c   | COCO-Stuff-171   |  7.1  |  11.0  |  24.8 |  53.3 |  92.6  |
+| LSeg+                              | Eff-B7   | COCO Panoptic    |  3.8  |   7.8  |  18.0 |  46.5 |    -   |
+| OpenSeg                            | Eff-B7   | COCO Panoptic    |  6.3  |   9.0  |  21.1 |  42.1 |    -   |
+| OVSeg (Ours)                       | Swin-B   | COCO-Stuff-171   |  9.0  |  12.4  |  29.6 |  55.7 |  94.5  |
+| Supervised specialist models.      |          |                  |       |        |       |       |        |
+| FCN                                | FCN-8s   | Same as test     |   -   |    -   |  29.4 |  37.8 |    -   |
+| Deeplab                            | R-101    | Same as test     |   -   |    -   |   -   |  45.7 |  77.7  |
+| SelfTrain                          | Eff-L2   | Same as test     |   -   |    -   |   -   |   -   |  90.0  |
+
+#### Ablation study
+
+- Mask prompt tuning can bring significant improvement without changing CLIP weights (Table 3 in [paper](https://arxiv.org/pdf/2210.04150.pdf))
+
+Download the checkpoint with mpt only [ovseg_swinbase_vitL14_mpt_only.pt](https://drive.google.com/file/d/1LJGWFjHw76OGDNy9r9KQIaACfIm9KMhQ/view?usp=sharing) (md5: <tt>2dd495</tt>).
+
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_mpt_only.pt DATASETS.TEST \(\"ade20k_sem_seg_val\",\"ade20k_full_sem_seg_val\"\) 
+  ```
+  
+- Mask prompt tuning can improve over fully finetuned model (Table 3 in [paper](https://arxiv.org/pdf/2210.04150.pdf))
+
+With the same [ovseg_swinbase_vitL14_ft_mpt.pth](https://drive.google.com/file/d/1cn-ohxgXDrDfkzC1QdO-fi8IjbjXmgKy/view?usp=sharing) checkpoint, set `MASK_PROMPT_FWD` as `False` 
+
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.CLIP_ADAPTER.MASK_PROMPT_FWD False MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth DATASETS.TEST \(\"ade20k_sem_seg_val\",\"ade20k_full_sem_seg_val\"\) 
+  ```
+
+- The effects of class prediction ensemble (Table 6 in [paper](https://arxiv.org/pdf/2210.04150.pdf))
+
+With the same [ovseg_swinbase_vitL14_ft_mpt.pth](https://drive.google.com/file/d/1cn-ohxgXDrDfkzC1QdO-fi8IjbjXmgKy/view?usp=sharing) checkpoint, set `CLIP_ENSEMBLE` as `False`.
+
+  ```bash
+  python train_net.py --num-gpu 8 --eval-only --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.CLIP_ADAPTER.CLIP_ENSEMBLE False MODEL.WEIGHTS #PATH_of_ovseg_swinbase_vitL14_ft_mpt.pth DATASETS.TEST \(\"ade20k_sem_seg_val\",\"ade20k_full_sem_seg_val\"\) 
+  ```
+
+### Training Segmentation model
+
+  Our model is trained on COCO-Stuff
+  
+- Training baseline w/ original CLIP
+  ```
+  python train_net.py --num-gpu 8 --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.CLIP_ADAPTER.MASK_PROMPT_FWD False
+  ```
+
+To reproduce our final results, you may want to use the our mask-adapted CLIP
+
+- Training ovseg w/ mask-adapted CLIP
+  ```
+  python train_net.py --num-gpu 8 --config-file configs/ovseg_swinB_vitL_bs32_120k.yaml MODEL.CLIP_ADAPTER.CLIP_MODEL_NAME #PATH_TO_MASKADAPTED_CLIP
+  ```
+  
+CAUTION: The final results is sensitive to the ensemble (appendix A.5 in [paper](https://arxiv.org/pdf/2210.04150.pdf)). Thus, you may want to use the ```tools/search_thr_ensemble_w.sh``` to find the best ensemble hyper-parameters.
+
+### Fine-tuning CLIP with collected mask-category pairs
+
+We are still working on this part, stay tuned!

INSTALL.md

@@ -0,0 +1,50 @@
+## Installation
+
+### Requirements
+- Linux with Python ≥ 3.8
+- PyTorch ≥ 1.8 and [torchvision](https://github.com/pytorch/vision/) that matches the PyTorch installation.
+  Install them together at [pytorch.org](https://pytorch.org) to make sure of this. Note, please check
+  PyTorch version matches that is required by Detectron2.
+- PyTorch3d: follow [Pytorch3d installation instructions](https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md).
+- Detectron2: follow [Detectron2 installation instructions](https://detectron2.readthedocs.io/tutorials/install.html).
+- Segment Anything Model: follow [SAM](https://github.com/facebookresearch/segment-anything).
+
+### Usage
+
+Install required packages. 
+
+```bash
+conda create --name ovseg python=3.8
+conda activate ovseg
+conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
+conda install -c fvcore -c iopath -c conda-forge fvcore iopath
+conda install pytorch3d -c pytorch3d
+pip install -r requirements.txt
+```
+
+You need to download `detectron2==0.6` following [instructions](https://detectron2.readthedocs.io/en/latest/tutorials/install.html)
+
+```bash
+python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html
+```
+
+If you cannot succefully install `pycocotools`, try this from [here](https://github.com/cocodataset/cocoapi/issues/351):
+```bash
+conda install -c conda-forge pycocotools
+```
+
+Install the SAM with:
+```bash
+pip install git+https://github.com/facebookresearch/segment-anything.git
+```
+To fully support the SAM, install these packages:
+```bash
+pip install opencv-python pycocotools matplotlib onnxruntime onnx
+```
+
+FurtherMore, install the modified clip package.
+
+```bash
+cd third_party/CLIP
+python -m pip install -Ue .
+```

LICENSE

@@ -0,0 +1,399 @@
+Attribution-NonCommercial 4.0 International
+
+=======================================================================
+
+Creative Commons Corporation ("Creative Commons") is not a law firm and
+does not provide legal services or legal advice. Distribution of
+Creative Commons public licenses does not create a lawyer-client or
+other relationship. Creative Commons makes its licenses and related
+information available on an "as-is" basis. Creative Commons gives no
+warranties regarding its licenses, any material licensed under their
+terms and conditions, or any related information. Creative Commons
+disclaims all liability for damages resulting from their use to the
+fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and
+conditions that creators and other rights holders may use to share
+original works of authorship and other material subject to copyright
+and certain other rights specified in the public license below. The
+following considerations are for informational purposes only, are not
+exhaustive, and do not form part of our licenses.
+
+     Considerations for licensors: Our public licenses are
+     intended for use by those authorized to give the public
+     permission to use material in ways otherwise restricted by
+     copyright and certain other rights. Our licenses are
+     irrevocable. Licensors should read and understand the terms
+     and conditions of the license they choose before applying it.
+     Licensors should also secure all rights necessary before
+     applying our licenses so that the public can reuse the
+     material as expected. Licensors should clearly mark any
+     material not subject to the license. This includes other CC-
+     licensed material, or material used under an exception or
+     limitation to copyright. More considerations for licensors:
+	wiki.creativecommons.org/Considerations_for_licensors
+
+     Considerations for the public: By using one of our public
+     licenses, a licensor grants the public permission to use the
+     licensed material under specified terms and conditions. If
+     the licensor's permission is not necessary for any reason--for
+     example, because of any applicable exception or limitation to
+     copyright--then that use is not regulated by the license. Our
+     licenses grant only permissions under copyright and certain
+     other rights that a licensor has authority to grant. Use of
+     the licensed material may still be restricted for other
+     reasons, including because others have copyright or other
+     rights in the material. A licensor may make special requests,
+     such as asking that all changes be marked or described.
+     Although not required by our licenses, you are encouraged to
+     respect those requests where reasonable. More_considerations
+     for the public:
+	wiki.creativecommons.org/Considerations_for_licensees
+
+=======================================================================
+
+Creative Commons Attribution-NonCommercial 4.0 International Public
+License
+
+By exercising the Licensed Rights (defined below), You accept and agree
+to be bound by the terms and conditions of this Creative Commons
+Attribution-NonCommercial 4.0 International Public License ("Public
+License"). To the extent this Public License may be interpreted as a
+contract, You are granted the Licensed Rights in consideration of Your
+acceptance of these terms and conditions, and the Licensor grants You
+such rights in consideration of benefits the Licensor receives from
+making the Licensed Material available under these terms and
+conditions.
+
+Section 1 -- Definitions.
+
+  a. Adapted Material means material subject to Copyright and Similar
+     Rights that is derived from or based upon the Licensed Material
+     and in which the Licensed Material is translated, altered,
+     arranged, transformed, or otherwise modified in a manner requiring
+     permission under the Copyright and Similar Rights held by the
+     Licensor. For purposes of this Public License, where the Licensed
+     Material is a musical work, performance, or sound recording,
+     Adapted Material is always produced where the Licensed Material is
+     synched in timed relation with a moving image.
+
+  b. Adapter's License means the license You apply to Your Copyright
+     and Similar Rights in Your contributions to Adapted Material in
+     accordance with the terms and conditions of this Public License.
+
+  c. Copyright and Similar Rights means copyright and/or similar rights
+     closely related to copyright including, without limitation,
+     performance, broadcast, sound recording, and Sui Generis Database
+     Rights, without regard to how the rights are labeled or
+     categorized. For purposes of this Public License, the rights
+     specified in Section 2(b)(1)-(2) are not Copyright and Similar
+     Rights.
+  d. Effective Technological Measures means those measures that, in the
+     absence of proper authority, may not be circumvented under laws
+     fulfilling obligations under Article 11 of the WIPO Copyright
+     Treaty adopted on December 20, 1996, and/or similar international
+     agreements.
+
+  e. Exceptions and Limitations means fair use, fair dealing, and/or
+     any other exception or limitation to Copyright and Similar Rights
+     that applies to Your use of the Licensed Material.
+
+  f. Licensed Material means the artistic or literary work, database,
+     or other material to which the Licensor applied this Public
+     License.
+
+  g. Licensed Rights means the rights granted to You subject to the
+     terms and conditions of this Public License, which are limited to
+     all Copyright and Similar Rights that apply to Your use of the
+     Licensed Material and that the Licensor has authority to license.
+
+  h. Licensor means the individual(s) or entity(ies) granting rights
+     under this Public License.
+
+  i. NonCommercial means not primarily intended for or directed towards
+     commercial advantage or monetary compensation. For purposes of
+     this Public License, the exchange of the Licensed Material for
+     other material subject to Copyright and Similar Rights by digital
+     file-sharing or similar means is NonCommercial provided there is
+     no payment of monetary compensation in connection with the
+     exchange.
+
+  j. Share means to provide material to the public by any means or
+     process that requires permission under the Licensed Rights, such
+     as reproduction, public display, public performance, distribution,
+     dissemination, communication, or importation, and to make material
+     available to the public including in ways that members of the
+     public may access the material from a place and at a time
+     individually chosen by them.
+
+  k. Sui Generis Database Rights means rights other than copyright
+     resulting from Directive 96/9/EC of the European Parliament and of
+     the Council of 11 March 1996 on the legal protection of databases,
+     as amended and/or succeeded, as well as other essentially
+     equivalent rights anywhere in the world.
+
+  l. You means the individual or entity exercising the Licensed Rights
+     under this Public License. Your has a corresponding meaning.
+
+Section 2 -- Scope.
+
+  a. License grant.
+
+       1. Subject to the terms and conditions of this Public License,
+          the Licensor hereby grants You a worldwide, royalty-free,
+          non-sublicensable, non-exclusive, irrevocable license to
+          exercise the Licensed Rights in the Licensed Material to:
+
+            a. reproduce and Share the Licensed Material, in whole or
+               in part, for NonCommercial purposes only; and
+
+            b. produce, reproduce, and Share Adapted Material for
+               NonCommercial purposes only.
+
+       2. Exceptions and Limitations. For the avoidance of doubt, where
+          Exceptions and Limitations apply to Your use, this Public
+          License does not apply, and You do not need to comply with
+          its terms and conditions.
+
+       3. Term. The term of this Public License is specified in Section
+          6(a).
+
+       4. Media and formats; technical modifications allowed. The
+          Licensor authorizes You to exercise the Licensed Rights in
+          all media and formats whether now known or hereafter created,
+          and to make technical modifications necessary to do so. The
+          Licensor waives and/or agrees not to assert any right or
+          authority to forbid You from making technical modifications
+          necessary to exercise the Licensed Rights, including
+          technical modifications necessary to circumvent Effective
+          Technological Measures. For purposes of this Public License,
+          simply making modifications authorized by this Section 2(a)
+          (4) never produces Adapted Material.
+
+       5. Downstream recipients.
+
+            a. Offer from the Licensor -- Licensed Material. Every
+               recipient of the Licensed Material automatically
+               receives an offer from the Licensor to exercise the
+               Licensed Rights under the terms and conditions of this
+               Public License.
+
+            b. No downstream restrictions. You may not offer or impose
+               any additional or different terms or conditions on, or
+               apply any Effective Technological Measures to, the
+               Licensed Material if doing so restricts exercise of the
+               Licensed Rights by any recipient of the Licensed
+               Material.
+
+       6. No endorsement. Nothing in this Public License constitutes or
+          may be construed as permission to assert or imply that You
+          are, or that Your use of the Licensed Material is, connected
+          with, or sponsored, endorsed, or granted official status by,
+          the Licensor or others designated to receive attribution as
+          provided in Section 3(a)(1)(A)(i).
+
+  b. Other rights.
+
+       1. Moral rights, such as the right of integrity, are not
+          licensed under this Public License, nor are publicity,
+          privacy, and/or other similar personality rights; however, to
+          the extent possible, the Licensor waives and/or agrees not to
+          assert any such rights held by the Licensor to the limited
+          extent necessary to allow You to exercise the Licensed
+          Rights, but not otherwise.
+
+       2. Patent and trademark rights are not licensed under this
+          Public License.
+
+       3. To the extent possible, the Licensor waives any right to
+          collect royalties from You for the exercise of the Licensed
+          Rights, whether directly or through a collecting society
+          under any voluntary or waivable statutory or compulsory
+          licensing scheme. In all other cases the Licensor expressly
+          reserves any right to collect such royalties, including when
+          the Licensed Material is used other than for NonCommercial
+          purposes.
+
+Section 3 -- License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the
+following conditions.
+
+  a. Attribution.
+
+       1. If You Share the Licensed Material (including in modified
+          form), You must:
+
+            a. retain the following if it is supplied by the Licensor
+               with the Licensed Material:
+
+                 i. identification of the creator(s) of the Licensed
+                    Material and any others designated to receive
+                    attribution, in any reasonable manner requested by
+                    the Licensor (including by pseudonym if
+                    designated);
+
+                ii. a copyright notice;
+
+               iii. a notice that refers to this Public License;
+
+                iv. a notice that refers to the disclaimer of
+                    warranties;
+
+                 v. a URI or hyperlink to the Licensed Material to the
+                    extent reasonably practicable;
+
+            b. indicate if You modified the Licensed Material and
+               retain an indication of any previous modifications; and
+
+            c. indicate the Licensed Material is licensed under this
+               Public License, and include the text of, or the URI or
+               hyperlink to, this Public License.
+
+       2. You may satisfy the conditions in Section 3(a)(1) in any
+          reasonable manner based on the medium, means, and context in
+          which You Share the Licensed Material. For example, it may be
+          reasonable to satisfy the conditions by providing a URI or
+          hyperlink to a resource that includes the required
+          information.
+
+       3. If requested by the Licensor, You must remove any of the
+          information required by Section 3(a)(1)(A) to the extent
+          reasonably practicable.
+
+       4. If You Share Adapted Material You produce, the Adapter's
+          License You apply must not prevent recipients of the Adapted
+          Material from complying with this Public License.
+
+Section 4 -- Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that
+apply to Your use of the Licensed Material:
+
+  a. for the avoidance of doubt, Section 2(a)(1) grants You the right
+     to extract, reuse, reproduce, and Share all or a substantial
+     portion of the contents of the database for NonCommercial purposes
+     only;
+
+  b. if You include all or a substantial portion of the database
+     contents in a database in which You have Sui Generis Database
+     Rights, then the database in which You have Sui Generis Database
+     Rights (but not its individual contents) is Adapted Material; and
+
+  c. You must comply with the conditions in Section 3(a) if You Share
+     all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not
+replace Your obligations under this Public License where the Licensed
+Rights include other Copyright and Similar Rights.
+
+Section 5 -- Disclaimer of Warranties and Limitation of Liability.
+
+  a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
+     EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
+     AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
+     ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
+     IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
+     WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
+     PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
+     ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
+     KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
+     ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
+
+  b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
+     TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
+     NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
+     INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
+     COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
+     USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
+     ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
+     DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
+     IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
+
+  c. The disclaimer of warranties and limitation of liability provided
+     above shall be interpreted in a manner that, to the extent
+     possible, most closely approximates an absolute disclaimer and
+     waiver of all liability.
+
+Section 6 -- Term and Termination.
+
+  a. This Public License applies for the term of the Copyright and
+     Similar Rights licensed here. However, if You fail to comply with
+     this Public License, then Your rights under this Public License
+     terminate automatically.
+
+  b. Where Your right to use the Licensed Material has terminated under
+     Section 6(a), it reinstates:
+
+       1. automatically as of the date the violation is cured, provided
+          it is cured within 30 days of Your discovery of the
+          violation; or
+
+       2. upon express reinstatement by the Licensor.
+
+     For the avoidance of doubt, this Section 6(b) does not affect any
+     right the Licensor may have to seek remedies for Your violations
+     of this Public License.
+
+  c. For the avoidance of doubt, the Licensor may also offer the
+     Licensed Material under separate terms or conditions or stop
+     distributing the Licensed Material at any time; however, doing so
+     will not terminate this Public License.
+
+  d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
+     License.
+
+Section 7 -- Other Terms and Conditions.
+
+  a. The Licensor shall not be bound by any additional or different
+     terms or conditions communicated by You unless expressly agreed.
+
+  b. Any arrangements, understandings, or agreements regarding the
+     Licensed Material not stated herein are separate from and
+     independent of the terms and conditions of this Public License.
+
+Section 8 -- Interpretation.
+
+  a. For the avoidance of doubt, this Public License does not, and
+     shall not be interpreted to, reduce, limit, restrict, or impose
+     conditions on any use of the Licensed Material that could lawfully
+     be made without permission under this Public License.
+
+  b. To the extent possible, if any provision of this Public License is
+     deemed unenforceable, it shall be automatically reformed to the
+     minimum extent necessary to make it enforceable. If the provision
+     cannot be reformed, it shall be severed from this Public License
+     without affecting the enforceability of the remaining terms and
+     conditions.
+
+  c. No term or condition of this Public License will be waived and no
+     failure to comply consented to unless expressly agreed to by the
+     Licensor.
+
+  d. Nothing in this Public License constitutes or may be interpreted
+     as a limitation upon, or waiver of, any privileges and immunities
+     that apply to the Licensor or You, including from the legal
+     processes of any jurisdiction or authority.
+
+=======================================================================
+
+Creative Commons is not a party to its public
+licenses. Notwithstanding, Creative Commons may elect to apply one of
+its public licenses to material it publishes and in those instances
+will be considered the “Licensor.” The text of the Creative Commons
+public licenses is dedicated to the public domain under the CC0 Public
+Domain Dedication. Except for the limited purpose of indicating that
+material is shared under a Creative Commons public license or as
+otherwise permitted by the Creative Commons policies published at
+creativecommons.org/policies, Creative Commons does not authorize the
+use of the trademark "Creative Commons" or any other trademark or logo
+of Creative Commons without its prior written consent including,
+without limitation, in connection with any unauthorized modifications
+to any of its public licenses or any other arrangements,
+understandings, or agreements concerning use of licensed material. For
+the avoidance of doubt, this paragraph does not form part of the
+public licenses.
+
+Creative Commons may be contacted at creativecommons.org.

UI/sailvos3d/ex1/inputs/depth_000160.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96d4969b8b33250785d1996b1536bb9026536f420391c68255e326990138598e
+size 4096128

UI/sailvos3d/ex1/inputs/rage_matrices_000160.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5afc2fdf1faa9b7b5d7808bb703c82aa5ccbb3154e2f62b3cc4989a2dcc92fe5
+size 1234

UI/sailvos3d/ex2/inputs/depth_000540.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f24fcabda3f7fd17856c4105279f2842b631cc18579d273b87dd8f2cb39e7df6
+size 4096128

UI/sailvos3d/ex2/inputs/rage_matrices_000540.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb66f05ce4cdb6d6410bd3e34b70eeb07724810e70786249c30de0f50404fd64
+size 1234

app.py

@@ -0,0 +1,194 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import argparse
+import glob
+import multiprocessing as mp
+import os
+import time
+import cv2
+import tqdm
+import numpy as np
+import gradio as gr
+
+from detectron2.config import get_cfg
+
+from detectron2.projects.deeplab import add_deeplab_config
+from detectron2.data.detection_utils import read_image
+from detectron2.utils.logger import setup_logger
+from open_vocab_seg import add_ovseg_config
+
+from open_vocab_seg.utils import VisualizationDemo
+
+# constants
+WINDOW_NAME = "Open vocabulary segmentation"
+
+
+def setup_cfg(args):
+    # load config from file and command-line arguments
+    cfg = get_cfg()
+    # for poly lr schedule
+    add_deeplab_config(cfg)
+    add_ovseg_config(cfg)
+    cfg.merge_from_file(args.config_file)
+    cfg.merge_from_list(args.opts)
+    cfg.freeze()
+    return cfg
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(description="Detectron2 demo for open vocabulary segmentation")
+    parser.add_argument(
+        "--config-file",
+        default="configs/ovseg_swinB_vitL_demo.yaml",
+        metavar="FILE",
+        help="path to config file",
+    )
+    parser.add_argument(
+        "--input",
+        default=["/mnt/lustre/jkyang/PSG4D/sailvos3d/downloads/sailvos3d/trevor_1_int/images/000160.bmp"],
+        nargs="+",
+        help="A list of space separated input images; "
+        "or a single glob pattern such as 'directory/*.jpg'",
+    )
+    parser.add_argument(
+        "--class-names",
+        default=["person", "car", "motorcycle", "truck", "bird", "dog", "handbag", "suitcase", "bottle", "cup", "bowl", "chair", "potted plant", "bed", "dining table", "tv", "laptop", "cell phone", "bag", "bin", "box", "door", "road barrier", "stick", "lamp", "floor", "wall"],
+        nargs="+",
+        help="A list of user-defined class_names"
+    )
+    parser.add_argument(
+        "--output", 
+        default = "./pred",
+        help="A file or directory to save output visualizations. "
+        "If not given, will show output in an OpenCV window.",
+    )
+    parser.add_argument(
+        "--opts",
+        help="Modify config options using the command-line 'KEY VALUE' pairs",
+        default=["MODEL.WEIGHTS", "ovseg_swinbase_vitL14_ft_mpt.pth"],
+        nargs=argparse.REMAINDER,
+    )
+    return parser
+
+args = get_parser().parse_args()
+
+def greet(rgb_input, depth_map_input, rage_matrices_input, class_candidates):
+    print(args.class_names)
+    print(class_candidates[0], class_candidates[1], class_candidates[2], class_candidates[3],)
+    print(class_candidates.split(', '))
+    args.input = [rgb_input]
+    args.class_names = class_candidates.split(', ')
+    depth_map_path = depth_map_input.name
+    rage_matrices_path = rage_matrices_input.name
+    print(args.input, args.class_names, depth_map_path, rage_matrices_path)
+    mp.set_start_method("spawn", force=True)
+    setup_logger(name="fvcore")
+    logger = setup_logger()
+    logger.info("Arguments: " + str(args))
+
+    cfg = setup_cfg(args)
+
+    demo = VisualizationDemo(cfg)
+    class_names = args.class_names
+    print(args.input)
+    if args.input:
+        if len(args.input) == 1:
+            args.input = glob.glob(os.path.expanduser(args.input[0]))
+            assert args.input, "The input path(s) was not found"
+        for path in tqdm.tqdm(args.input, disable=not args.output):
+            # use PIL, to be consistent with evaluation
+            start_time = time.time()
+            predictions, visualized_output_rgb, visualized_output_depth, visualized_output_rgb_sam, visualized_output_depth_sam = demo.run_on_image_sam(path, class_names, depth_map_path, rage_matrices_path)
+            logger.info(
+                "{}: {} in {:.2f}s".format(
+                    path,
+                    "detected {} instances".format(len(predictions["instances"]))
+                    if "instances" in predictions
+                    else "finished",
+                    time.time() - start_time,
+                )
+            )
+
+            if args.output:
+                if os.path.isdir(args.output):
+                    assert os.path.isdir(args.output), args.output
+                    out_filename = os.path.join(args.output, os.path.basename(path))
+                else:
+                    assert len(args.input) == 1, "Please specify a directory with args.output"
+                    out_filename = args.output
+                visualized_output_rgb.save('outputs/RGB_Semantic_SAM.png')
+                visualized_output_depth.save('outputs/Depth_Semantic_SAM.png')
+                visualized_output_rgb_sam.save('outputs/RGB_Semantic_SAM_Mask.png')
+                visualized_output_depth_sam.save('outputs/Depth_Semantic_SAM_Mask.png')
+                rgb_3d_sam = demo.get_xyzrgb('outputs/RGB_Semantic_SAM.png', depth_map_path, rage_matrices_path)
+                depth_3d_sam = demo.get_xyzrgb('outputs/Depth_Semantic_SAM.png', depth_map_path, rage_matrices_path)
+                rgb_3d_sam_mask = demo.get_xyzrgb('outputs/RGB_Semantic_SAM_Mask.png', depth_map_path, rage_matrices_path)
+                depth_3d_sam_mask = demo.get_xyzrgb('outputs/Depth_Semantic_SAM_Mask.png', depth_map_path, rage_matrices_path)
+                np.savez('outputs/xyzrgb.npz', rgb_3d_sam = rgb_3d_sam, depth_3d_sam = depth_3d_sam, rgb_3d_sam_mask = rgb_3d_sam_mask, depth_3d_sam_mask = depth_3d_sam_mask)
+                demo.render_3d_video('outputs/xyzrgb.npz', depth_map_path)
+            else:
+                cv2.namedWindow(WINDOW_NAME, cv2.WINDOW_NORMAL)
+                cv2.imshow(WINDOW_NAME, visualized_output_rgb.get_image()[:, :, ::-1])
+                if cv2.waitKey(0) == 27:
+                    break  # esc to quit
+    else:
+        raise NotImplementedError
+    
+    Depth_Semantic_SAM_Mask = read_image('outputs/Depth_Semantic_SAM_Mask.png')
+    RGB_Semantic_SAM_Mask = read_image('outputs/RGB_Semantic_SAM_Mask.png')
+    Depth_map = read_image('outputs/Depth_rendered.png')
+    Depth_Semantic_SAM_Mask_gif = 'outputs/depth_3d_sam_mask.gif'
+    RGB_Semantic_SAM_Mask_gif = 'outputs/rgb_3d_sam_mask.gif'
+    return RGB_Semantic_SAM_Mask, RGB_Semantic_SAM_Mask_gif, Depth_map, Depth_Semantic_SAM_Mask, Depth_Semantic_SAM_Mask_gif
+
+with gr.Blocks(analytics_enabled=False) as segrgbd_iface:
+        gr.Markdown("<div align='center'> <h2> Semantic Segment AnyRGBD </span> </h2> \
+                     <a style='font-size:18px;color: #000000' href='https://github.com/Jun-CEN/SegmentAnyRGBD'> Github </div>")
+        
+        gr.Markdown("<b> You may duplicate the space and upgrade to GPU in settings for better performance and faster inference without waiting in the queue. <a style='display:inline-block' href='https://huggingface.co/spaces/VideoCrafter/VideoCrafter?duplicate=true'> <img src='https://bit.ly/3gLdBN6' alt='Duplicate Space'></a> </b>")
+        #######t2v#######
+        with gr.Tab(label="Dataset: Sailvos3D"):
+            with gr.Column():
+                with gr.Row():
+                    # with gr.Tab(label='input'):
+                    with gr.Column():
+                        with gr.Row():
+                            Input_RGB_Component = gr.Image(label = 'RGB_Input', type = 'filepath').style(width=320, height=200)
+                            Depth_Map_Output_Component = gr.Image(label = "Depth_Map").style(width=320, height=200)
+                        with gr.Row():
+                            Depth_Map_Input_Component = gr.File(label = 'Depth_map')
+                            Component_2D_to_3D_Projection_Parameters = gr.File(label = '2D_to_3D_Projection_Parameters')
+                        with gr.Row():
+                            Class_Candidates_Component = gr.Text(label = 'Class_Candidates')
+                        vc_end_btn = gr.Button("Send")
+                    with gr.Tab(label='Result'):
+                        with gr.Row():
+                            RGB_Semantic_SAM_Mask_Component = gr.Image(label = "RGB_Semantic_SAM_Mask").style(width=320, height=200)
+                            RGB_Semantic_SAM_Mask_3D_Component = gr.Image(label = "3D_RGB_Semantic_SAM_Mask").style(width=320, height=200)
+                        with gr.Row():
+                            Depth_Semantic_SAM_Mask_Component = gr.Image(label = "Depth_Semantic_SAM_Mask").style(width=320, height=200)
+                            Depth_Semantic_SAM_Mask_3D_Component = gr.Image(label = "3D_Depth_Semantic_SAM_Mask").style(width=320, height=200)
+                gr.Examples(examples=[
+                        [
+                            'UI/sailvos3d/ex1/inputs/rgb_000160.bmp',
+                            'UI/sailvos3d/ex1/inputs/depth_000160.npy',
+                            'UI/sailvos3d/ex1/inputs/rage_matrices_000160.npz',
+                            'person, car, motorcycle, truck, bird, dog, handbag, suitcase, bottle, cup, bowl, chair, potted plant, bed, dining table, tv, laptop, cell phone, bag, bin, box, door, road barrier, stick, lamp, floor, wall',
+                        ],
+                        [
+                            'UI/sailvos3d/ex2/inputs/rgb_000540.bmp',
+                            'UI/sailvos3d/ex2/inputs/depth_000540.npy',
+                            'UI/sailvos3d/ex2/inputs/rage_matrices_000540.npz',
+                            'person, car, motorcycle, truck, bird, dog, handbag, suitcase, bottle, cup, bowl, chair, potted plant, bed, dining table, tv, laptop, cell phone, bag, bin, box, door, road barrier, stick, lamp, floor, wall',
+                        ]],
+                            inputs=[Input_RGB_Component, Depth_Map_Input_Component, Component_2D_to_3D_Projection_Parameters, Class_Candidates_Component],
+                            outputs=[RGB_Semantic_SAM_Mask_Component, RGB_Semantic_SAM_Mask_3D_Component, Depth_Map_Output_Component, Depth_Semantic_SAM_Mask_Component, Depth_Semantic_SAM_Mask_3D_Component],
+                            fn=greet)
+            vc_end_btn.click(inputs=[Input_RGB_Component, Depth_Map_Input_Component, Component_2D_to_3D_Projection_Parameters, Class_Candidates_Component],
+                            outputs=[RGB_Semantic_SAM_Mask_Component, RGB_Semantic_SAM_Mask_3D_Component, Depth_Map_Output_Component, Depth_Semantic_SAM_Mask_Component, Depth_Semantic_SAM_Mask_3D_Component],
+                            fn=greet)
+
+demo = segrgbd_iface
+demo.launch()
+

configs/ovseg_swinB_vitL_bs32_120k.yaml

@@ -0,0 +1,100 @@
+MODEL:
+  META_ARCHITECTURE: "OVSeg"
+  BACKBONE:
+    FREEZE_AT: 0
+    NAME: "D2SwinTransformer"
+  SWIN:
+    EMBED_DIM: 128
+    DEPTHS: [2, 2, 18, 2]
+    NUM_HEADS: [4, 8, 16, 32]
+    WINDOW_SIZE: 12
+    APE: False
+    DROP_PATH_RATE: 0.3
+    PATCH_NORM: True
+    PRETRAIN_IMG_SIZE: 384
+  WEIGHTS: "swin_base_patch4_window12_384_22k.pkl"
+  PIXEL_MEAN: [123.675, 116.280, 103.530]
+  PIXEL_STD: [58.395, 57.120, 57.375]
+  SEM_SEG_HEAD:
+    NAME: "OpenVocabMaskFormerHead"
+    IN_FEATURES: ["res2", "res3", "res4", "res5"]
+    IGNORE_VALUE: 255
+    NUM_CLASSES: 171 # number of categories in training set
+    EMBEDDING_DIM: 768
+    EMBED_LAYERS: 2
+    COMMON_STRIDE: 4 # not used, hard-coded
+    LOSS_WEIGHT: 1.0
+    CONVS_DIM: 256
+    MASK_DIM: 256
+    NORM: "GN"
+  MASK_FORMER:
+    TRANSFORMER_IN_FEATURE: "res5"
+    DEEP_SUPERVISION: True
+    NO_OBJECT_WEIGHT: 0.1
+    DICE_WEIGHT: 1.0
+    MASK_WEIGHT: 20.0
+    HIDDEN_DIM: 256
+    NUM_OBJECT_QUERIES: 100
+    NHEADS: 8
+    DROPOUT: 0.1
+    DIM_FEEDFORWARD: 2048
+    ENC_LAYERS: 0
+    DEC_LAYERS: 6
+    PRE_NORM: False
+  CLIP_ADAPTER:
+    TEXT_TEMPLATES: "vild"
+    CLIP_MODEL_NAME: "ViT-L/14"
+    MASK_FILL: "mean"
+    MASK_EXPAND_RATIO: 1.0
+    MASK_THR: 0.4 # choose the foreground objects
+    MASK_MATTING: False # use soft background, default not used
+    MASK_PROMPT_DEPTH: 3
+    MASK_PROMPT_FWD: True # use mask prompt during forward
+    REGION_RESIZED: True # resize to the input of clip, e.g., 224
+    CLIP_ENSEMBLE: True # use ensemble of two classification branches
+    CLIP_ENSEMBLE_WEIGHT: 0.7
+DATASETS:
+  TRAIN: ("coco_2017_train_stuff_sem_seg",)
+  TEST: ("ade20k_sem_seg_val",)
+SOLVER:
+  IMS_PER_BATCH: 32
+  BASE_LR: 0.00006
+  MAX_ITER: 120000
+  WARMUP_FACTOR: 1e-6
+  WARMUP_ITERS: 1500
+  LR_SCHEDULER_NAME: "WarmupPolyLR"
+  WEIGHT_DECAY: 0.01
+  WEIGHT_DECAY_NORM: 0.0
+  WEIGHT_DECAY_EMBED: 0.0
+  BACKBONE_MULTIPLIER: 1.0
+  TEST_IMS_PER_BATCH: 1
+  CLIP_GRADIENTS:
+    ENABLED: True
+    CLIP_TYPE: "full_model"
+    CLIP_VALUE: 0.01
+    NORM_TYPE: 2.0
+INPUT:
+  MIN_SIZE_TRAIN: !!python/object/apply:eval ["[int(x * 0.1 * 640) for x in range(5, 21)]"]
+  MIN_SIZE_TRAIN_SAMPLING: "choice"
+  MIN_SIZE_TEST: 640
+  MAX_SIZE_TRAIN: 2560
+  MAX_SIZE_TEST: 2560
+  CROP:
+    ENABLED: True
+    TYPE: "absolute"
+    SIZE: (640, 640)
+    SINGLE_CATEGORY_MAX_AREA: 1.0
+  COLOR_AUG_SSD: True
+  SIZE_DIVISIBILITY: 640  # used in dataset mapper
+  FORMAT: "RGB"
+TEST:
+  EVAL_PERIOD: 5000
+  AUG:
+    ENABLED: False
+    MIN_SIZES: [256, 384, 512, 640, 768, 896]
+    MAX_SIZE: 3584
+    FLIP: True
+DATALOADER:
+  FILTER_EMPTY_ANNOTATIONS: True
+  NUM_WORKERS: 4
+VERSION: 2

configs/ovseg_swinB_vitL_demo.yaml

@@ -0,0 +1,99 @@
+MODEL:
+  META_ARCHITECTURE: "OVSegDEMO"
+  BACKBONE:
+    FREEZE_AT: 0
+    NAME: "D2SwinTransformer"
+  SWIN:
+    EMBED_DIM: 128
+    DEPTHS: [2, 2, 18, 2]
+    NUM_HEADS: [4, 8, 16, 32]
+    WINDOW_SIZE: 12
+    APE: False
+    DROP_PATH_RATE: 0.3
+    PATCH_NORM: True
+    PRETRAIN_IMG_SIZE: 384
+  WEIGHTS: "swin_base_patch4_window12_384_22k.pkl"
+  PIXEL_MEAN: [123.675, 116.280, 103.530]
+  PIXEL_STD: [58.395, 57.120, 57.375]
+  SEM_SEG_HEAD:
+    NAME: "OpenVocabMaskFormerHead"
+    IN_FEATURES: ["res2", "res3", "res4", "res5"]
+    IGNORE_VALUE: 255
+    NUM_CLASSES: 171 # number of categories in training set
+    EMBEDDING_DIM: 768
+    EMBED_LAYERS: 2
+    COMMON_STRIDE: 4 # not used, hard-coded
+    LOSS_WEIGHT: 1.0
+    CONVS_DIM: 256
+    MASK_DIM: 256
+    NORM: "GN"
+  MASK_FORMER:
+    TRANSFORMER_IN_FEATURE: "res5"
+    DEEP_SUPERVISION: True
+    NO_OBJECT_WEIGHT: 0.1
+    DICE_WEIGHT: 1.0
+    MASK_WEIGHT: 20.0
+    HIDDEN_DIM: 256
+    NUM_OBJECT_QUERIES: 100
+    NHEADS: 8
+    DROPOUT: 0.1
+    DIM_FEEDFORWARD: 2048
+    ENC_LAYERS: 0
+    DEC_LAYERS: 6
+    PRE_NORM: False
+  CLIP_ADAPTER:
+    TEXT_TEMPLATES: "vild"
+    CLIP_MODEL_NAME: "ViT-L/14"
+    MASK_FILL: "mean"
+    MASK_EXPAND_RATIO: 1.0
+    MASK_THR: 0.1 # choose the foreground objects
+    MASK_MATTING: False # use soft background, default not used
+    MASK_PROMPT_DEPTH: 3
+    MASK_PROMPT_FWD: True # use mask prompt during forward
+    REGION_RESIZED: True # resize to the input of clip, e.g., 224
+    CLIP_ENSEMBLE: True # use ensemble of two classification branches
+    CLIP_ENSEMBLE_WEIGHT: 0.0
+DATASETS:
+  TRAIN: ("coco_2017_train_stuff_sem_seg",)
+  TEST: ("ade20k_sem_seg_val",)
+SOLVER:
+  IMS_PER_BATCH: 32
+  BASE_LR: 0.00006
+  MAX_ITER: 120000
+  WARMUP_FACTOR: 1e-6
+  WARMUP_ITERS: 1500
+  WEIGHT_DECAY: 0.01
+  WEIGHT_DECAY_NORM: 0.0
+  WEIGHT_DECAY_EMBED: 0.0
+  BACKBONE_MULTIPLIER: 1.0
+  TEST_IMS_PER_BATCH: 1
+  CLIP_GRADIENTS:
+    ENABLED: True
+    CLIP_TYPE: "full_model"
+    CLIP_VALUE: 0.01
+    NORM_TYPE: 2.0
+INPUT:
+  MIN_SIZE_TRAIN: !!python/object/apply:eval ["[int(x * 0.1 * 640) for x in range(5, 21)]"]
+  MIN_SIZE_TRAIN_SAMPLING: "choice"
+  MIN_SIZE_TEST: 640
+  MAX_SIZE_TRAIN: 2560
+  MAX_SIZE_TEST: 2560
+  CROP:
+    ENABLED: True
+    TYPE: "absolute"
+    SIZE: (640, 640)
+    SINGLE_CATEGORY_MAX_AREA: 1.0
+  COLOR_AUG_SSD: True
+  SIZE_DIVISIBILITY: 640  # used in dataset mapper
+  FORMAT: "RGB"
+TEST:
+  EVAL_PERIOD: 5000
+  AUG:
+    ENABLED: False
+    MIN_SIZES: [256, 384, 512, 640, 768, 896]
+    MAX_SIZE: 3584
+    FLIP: True
+DATALOADER:
+  FILTER_EMPTY_ANNOTATIONS: True
+  NUM_WORKERS: 4
+VERSION: 2

datasets/DATASETS.md

@@ -0,0 +1,122 @@
+## Prepare Datasets for OVSeg
+
+This doc is a modification/extension of [MaskFormer](https://github.com/facebookresearch/MaskFormer/blob/main/datasets/README.md) following [Detectron2 fromat](https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html).
+
+A dataset can be used by accessing [DatasetCatalog](https://detectron2.readthedocs.io/modules/data.html#detectron2.data.DatasetCatalog)
+for its data, or [MetadataCatalog](https://detectron2.readthedocs.io/modules/data.html#detectron2.data.MetadataCatalog) for its metadata (class names, etc).
+This document explains how to setup the builtin datasets so they can be used by the above APIs.
+[Use Custom Datasets](https://detectron2.readthedocs.io/tutorials/datasets.html) gives a deeper dive on how to use `DatasetCatalog` and `MetadataCatalog`,
+and how to add new datasets to them.
+
+OVSeg has builtin support for a few datasets.
+The datasets are assumed to exist in a directory specified by the environment variable
+`DETECTRON2_DATASETS`.
+Under this directory, detectron2 will look for datasets in the structure described below, if needed.
+```
+$DETECTRON2_DATASETS/
+  coco/                 # COCOStuff-171
+  ADEChallengeData2016/ # ADE20K-150
+  ADE20K_2021_17_01/    # ADE20K-847
+  VOCdevkit/
+    VOC2012/            # PASCALVOC-20
+    VOC2010/            # PASCALContext-59, PASCALContext-459
+```
+
+You can set the location for builtin datasets by `export DETECTRON2_DATASETS=/path/to/datasets`.
+If left unset, the default is `./datasets` relative to your current working directory.
+
+Without specific notifications, our model is trained on COCOStuff-171 and evlauted on ADE20K-150, ADE20K-847, PASCALVOC-20, PASCALContext-59 and PASCALContext-459.
+
+|     dataset    |   split   | # images | # categories |
+|:--------------:|:---------:|:--------:|:------------:|
+|   COCO Stuff   | train2017 |   118K   |      171     |
+|     ADE20K     |    val    |    2K    |    150/847   |
+|   Pascal VOC   |    val    |   1.5K   |      20      |
+| Pascal Context |    val    |    5K    |    59/459    |
+
+
+### Expected dataset structure for [COCO Stuff](https://github.com/nightrome/cocostuff):
+```
+coco/
+  train2017/ # http://images.cocodataset.org/zips/train2017.zip
+  annotations/ # http://images.cocodataset.org/annotations/annotations_trainval2017.zip
+  stuffthingmaps/
+    stuffthingmaps_trainval2017.zip # http://calvin.inf.ed.ac.uk/wp-content/uploads/data/cocostuffdataset/stuffthingmaps_trainval2017.zip
+    train2017/
+  # below are generated
+  stuffthingmaps_detectron2/ 
+    train2017/
+```
+
+The directory `stuffthingmaps_detectron2` is generated by running `python datasets/prepare_coco_stuff_sem_seg.py`.
+
+
+
+### Expected dataset structure for [ADE20k Scene Parsing (ADE20K-150)](http://sceneparsing.csail.mit.edu/):
+```
+ADEChallengeData2016/
+  annotations/
+  images/
+  objectInfo150.txt
+  # below are generated
+  annotations_detectron2/
+```
+The directory `annotations_detectron2` is generated by running `python datasets/prepare_ade20k_sem_seg.py`.
+
+
+### Expected dataset structure for [ADE20k-Full (ADE20K-847)](https://github.com/CSAILVision/ADE20K#download):
+```
+ADE20K_2021_17_01/
+  images/
+  index_ade20k.pkl
+  objects.txt
+  # below are generated
+  images_detectron2/
+  annotations_detectron2/
+```
+The directories `images_detectron2` and `annotations_detectron2` are generated by running `python datasets/prepare_ade20k_full_sem_seg.py`.
+
+### Expected dataset structure for [Pascal VOC 2012 (PASCALVOC-20)](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/#devkit):
+```
+VOCdevkit/VOC2012/
+  Annotations/
+  ImageSets/
+  JPEGImages/
+  SegmentationClass/
+  SegmentationObject/
+  SegmentationClassAug/ # https://github.com/kazuto1011/deeplab-pytorch/blob/master/data/datasets/voc12/README.md
+  # below are generated
+  images_detectron2/
+  annotations_detectron2/
+```
+
+It starts with a tar file `VOCtrainval_11-May-2012.tar`.
+
+We use SBD augmentated training data as `SegmentationClassAug` following [Deeplab](https://github.com/kazuto1011/deeplab-pytorch/blob/master/data/datasets/voc12/README.md)
+
+The directories `images_detectron2` and `annotations_detectron2` are generated by running `python datasets/prepare_voc_sem_seg.py`.
+
+
+### Expected dataset structure for [Pascal Context](https://www.cs.stanford.edu/~roozbeh/pascal-context/):
+
+```
+VOCdevkit/VOC2010/
+  Annotations/
+  ImageSets/
+  JPEGImages/
+  SegmentationClass/
+  SegmentationObject/
+  # below are from https://www.cs.stanford.edu/~roozbeh/pascal-context/trainval.tar.gz
+  trainval/
+  labels.txt
+  59_labels.txt # https://www.cs.stanford.edu/~roozbeh/pascal-context/59_labels.txt
+  pascalcontext_val.txt # https://drive.google.com/file/d/1BCbiOKtLvozjVnlTJX51koIveUZHCcUh/view?usp=sharing
+  # below are generated
+  annotations_detectron2/
+    pc459_val
+    pc59_val
+```
+It starts with a tar file `VOCtrainval_03-May-2010.tar`. You may want to download the 5K validation set [here](https://drive.google.com/file/d/1BCbiOKtLvozjVnlTJX51koIveUZHCcUh/view?usp=sharing).
+
+The directory `annotations_detectron2` is generated by running `python datasets/prepare_pascal_context.py`.
+

datasets/prepare_ade20k_full_sem_seg.py

@@ -0,0 +1,1011 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import os
+import pickle as pkl
+from pathlib import Path
+
+import cv2
+import numpy as np
+import tqdm
+from PIL import Image
+
+ADE20K_SEM_SEG_FULL_CATEGORIES = [
+    {"name": "wall", "id": 2978, "trainId": 0},
+    {"name": "building, edifice", "id": 312, "trainId": 1},
+    {"name": "sky", "id": 2420, "trainId": 2},
+    {"name": "tree", "id": 2855, "trainId": 3},
+    {"name": "road, route", "id": 2131, "trainId": 4},
+    {"name": "floor, flooring", "id": 976, "trainId": 5},
+    {"name": "ceiling", "id": 447, "trainId": 6},
+    {"name": "bed", "id": 165, "trainId": 7},
+    {"name": "sidewalk, pavement", "id": 2377, "trainId": 8},
+    {"name": "earth, ground", "id": 838, "trainId": 9},
+    {"name": "cabinet", "id": 350, "trainId": 10},
+    {"name": "person, individual, someone, somebody, mortal, soul", "id": 1831, "trainId": 11},
+    {"name": "grass", "id": 1125, "trainId": 12},
+    {"name": "windowpane, window", "id": 3055, "trainId": 13},
+    {"name": "car, auto, automobile, machine, motorcar", "id": 401, "trainId": 14},
+    {"name": "mountain, mount", "id": 1610, "trainId": 15},
+    {"name": "plant, flora, plant life", "id": 1910, "trainId": 16},
+    {"name": "table", "id": 2684, "trainId": 17},
+    {"name": "chair", "id": 471, "trainId": 18},
+    {"name": "curtain, drape, drapery, mantle, pall", "id": 687, "trainId": 19},
+    {"name": "door", "id": 774, "trainId": 20},
+    {"name": "sofa, couch, lounge", "id": 2473, "trainId": 21},
+    {"name": "sea", "id": 2264, "trainId": 22},
+    {"name": "painting, picture", "id": 1735, "trainId": 23},
+    {"name": "water", "id": 2994, "trainId": 24},
+    {"name": "mirror", "id": 1564, "trainId": 25},
+    {"name": "house", "id": 1276, "trainId": 26},
+    {"name": "rug, carpet, carpeting", "id": 2178, "trainId": 27},
+    {"name": "shelf", "id": 2329, "trainId": 28},
+    {"name": "armchair", "id": 57, "trainId": 29},
+    {"name": "fence, fencing", "id": 907, "trainId": 30},
+    {"name": "field", "id": 913, "trainId": 31},
+    {"name": "lamp", "id": 1395, "trainId": 32},
+    {"name": "rock, stone", "id": 2138, "trainId": 33},
+    {"name": "seat", "id": 2272, "trainId": 34},
+    {"name": "river", "id": 2128, "trainId": 35},
+    {"name": "desk", "id": 724, "trainId": 36},
+    {"name": "bathtub, bathing tub, bath, tub", "id": 155, "trainId": 37},
+    {"name": "railing, rail", "id": 2053, "trainId": 38},
+    {"name": "signboard, sign", "id": 2380, "trainId": 39},
+    {"name": "cushion", "id": 689, "trainId": 40},
+    {"name": "path", "id": 1788, "trainId": 41},
+    {"name": "work surface", "id": 3087, "trainId": 42},
+    {"name": "stairs, steps", "id": 2530, "trainId": 43},
+    {"name": "column, pillar", "id": 581, "trainId": 44},
+    {"name": "sink", "id": 2388, "trainId": 45},
+    {"name": "wardrobe, closet, press", "id": 2985, "trainId": 46},
+    {"name": "snow", "id": 2454, "trainId": 47},
+    {"name": "refrigerator, icebox", "id": 2096, "trainId": 48},
+    {"name": "base, pedestal, stand", "id": 137, "trainId": 49},
+    {"name": "bridge, span", "id": 294, "trainId": 50},
+    {"name": "blind, screen", "id": 212, "trainId": 51},
+    {"name": "runway", "id": 2185, "trainId": 52},
+    {"name": "cliff, drop, drop-off", "id": 524, "trainId": 53},
+    {"name": "sand", "id": 2212, "trainId": 54},
+    {"name": "fireplace, hearth, open fireplace", "id": 943, "trainId": 55},
+    {"name": "pillow", "id": 1869, "trainId": 56},
+    {"name": "screen door, screen", "id": 2251, "trainId": 57},
+    {"name": "toilet, can, commode, crapper, pot, potty, stool, throne", "id": 2793, "trainId": 58},
+    {"name": "skyscraper", "id": 2423, "trainId": 59},
+    {"name": "grandstand, covered stand", "id": 1121, "trainId": 60},
+    {"name": "box", "id": 266, "trainId": 61},
+    {"name": "pool table, billiard table, snooker table", "id": 1948, "trainId": 62},
+    {"name": "palm, palm tree", "id": 1744, "trainId": 63},
+    {"name": "double door", "id": 783, "trainId": 64},
+    {"name": "coffee table, cocktail table", "id": 571, "trainId": 65},
+    {"name": "counter", "id": 627, "trainId": 66},
+    {"name": "countertop", "id": 629, "trainId": 67},
+    {"name": "chest of drawers, chest, bureau, dresser", "id": 491, "trainId": 68},
+    {"name": "kitchen island", "id": 1374, "trainId": 69},
+    {"name": "boat", "id": 223, "trainId": 70},
+    {"name": "waterfall, falls", "id": 3016, "trainId": 71},
+    {
+        "name": "stove, kitchen stove, range, kitchen range, cooking stove",
+        "id": 2598,
+        "trainId": 72,
+    },
+    {"name": "flower", "id": 978, "trainId": 73},
+    {"name": "bookcase", "id": 239, "trainId": 74},
+    {"name": "controls", "id": 608, "trainId": 75},
+    {"name": "book", "id": 236, "trainId": 76},
+    {"name": "stairway, staircase", "id": 2531, "trainId": 77},
+    {"name": "streetlight, street lamp", "id": 2616, "trainId": 78},
+    {
+        "name": "computer, computing machine, computing device, data processor, electronic computer, information processing system",
+        "id": 591,
+        "trainId": 79,
+    },
+    {
+        "name": "bus, autobus, coach, charabanc, double-decker, jitney, motorbus, motorcoach, omnibus, passenger vehicle",
+        "id": 327,
+        "trainId": 80,
+    },
+    {"name": "swivel chair", "id": 2679, "trainId": 81},
+    {"name": "light, light source", "id": 1451, "trainId": 82},
+    {"name": "bench", "id": 181, "trainId": 83},
+    {"name": "case, display case, showcase, vitrine", "id": 420, "trainId": 84},
+    {"name": "towel", "id": 2821, "trainId": 85},
+    {"name": "fountain", "id": 1023, "trainId": 86},
+    {"name": "embankment", "id": 855, "trainId": 87},
+    {
+        "name": "television receiver, television, television set, tv, tv set, idiot box, boob tube, telly, goggle box",
+        "id": 2733,
+        "trainId": 88,
+    },
+    {"name": "van", "id": 2928, "trainId": 89},
+    {"name": "hill", "id": 1240, "trainId": 90},
+    {"name": "awning, sunshade, sunblind", "id": 77, "trainId": 91},
+    {"name": "poster, posting, placard, notice, bill, card", "id": 1969, "trainId": 92},
+    {"name": "truck, motortruck", "id": 2880, "trainId": 93},
+    {"name": "airplane, aeroplane, plane", "id": 14, "trainId": 94},
+    {"name": "pole", "id": 1936, "trainId": 95},
+    {"name": "tower", "id": 2828, "trainId": 96},
+    {"name": "court", "id": 631, "trainId": 97},
+    {"name": "ball", "id": 103, "trainId": 98},
+    {
+        "name": "aircraft carrier, carrier, flattop, attack aircraft carrier",
+        "id": 3144,
+        "trainId": 99,
+    },
+    {"name": "buffet, counter, sideboard", "id": 308, "trainId": 100},
+    {"name": "hovel, hut, hutch, shack, shanty", "id": 1282, "trainId": 101},
+    {"name": "apparel, wearing apparel, dress, clothes", "id": 38, "trainId": 102},
+    {"name": "minibike, motorbike", "id": 1563, "trainId": 103},
+    {"name": "animal, animate being, beast, brute, creature, fauna", "id": 29, "trainId": 104},
+    {"name": "chandelier, pendant, pendent", "id": 480, "trainId": 105},
+    {"name": "step, stair", "id": 2569, "trainId": 106},
+    {"name": "booth, cubicle, stall, kiosk", "id": 247, "trainId": 107},
+    {"name": "bicycle, bike, wheel, cycle", "id": 187, "trainId": 108},
+    {"name": "doorframe, doorcase", "id": 778, "trainId": 109},
+    {"name": "sconce", "id": 2243, "trainId": 110},
+    {"name": "pond", "id": 1941, "trainId": 111},
+    {"name": "trade name, brand name, brand, marque", "id": 2833, "trainId": 112},
+    {"name": "bannister, banister, balustrade, balusters, handrail", "id": 120, "trainId": 113},
+    {"name": "bag", "id": 95, "trainId": 114},
+    {"name": "traffic light, traffic signal, stoplight", "id": 2836, "trainId": 115},
+    {"name": "gazebo", "id": 1087, "trainId": 116},
+    {"name": "escalator, moving staircase, moving stairway", "id": 868, "trainId": 117},
+    {"name": "land, ground, soil", "id": 1401, "trainId": 118},
+    {"name": "board, plank", "id": 220, "trainId": 119},
+    {"name": "arcade machine", "id": 47, "trainId": 120},
+    {"name": "eiderdown, duvet, continental quilt", "id": 843, "trainId": 121},
+    {"name": "bar", "id": 123, "trainId": 122},
+    {"name": "stall, stand, sales booth", "id": 2537, "trainId": 123},
+    {"name": "playground", "id": 1927, "trainId": 124},
+    {"name": "ship", "id": 2337, "trainId": 125},
+    {"name": "ottoman, pouf, pouffe, puff, hassock", "id": 1702, "trainId": 126},
+    {
+        "name": "ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin",
+        "id": 64,
+        "trainId": 127,
+    },
+    {"name": "bottle", "id": 249, "trainId": 128},
+    {"name": "cradle", "id": 642, "trainId": 129},
+    {"name": "pot, flowerpot", "id": 1981, "trainId": 130},
+    {
+        "name": "conveyer belt, conveyor belt, conveyer, conveyor, transporter",
+        "id": 609,
+        "trainId": 131,
+    },
+    {"name": "train, railroad train", "id": 2840, "trainId": 132},
+    {"name": "stool", "id": 2586, "trainId": 133},
+    {"name": "lake", "id": 1393, "trainId": 134},
+    {"name": "tank, storage tank", "id": 2704, "trainId": 135},
+    {"name": "ice, water ice", "id": 1304, "trainId": 136},
+    {"name": "basket, handbasket", "id": 146, "trainId": 137},
+    {"name": "manhole", "id": 1494, "trainId": 138},
+    {"name": "tent, collapsible shelter", "id": 2739, "trainId": 139},
+    {"name": "canopy", "id": 389, "trainId": 140},
+    {"name": "microwave, microwave oven", "id": 1551, "trainId": 141},
+    {"name": "barrel, cask", "id": 131, "trainId": 142},
+    {"name": "dirt track", "id": 738, "trainId": 143},
+    {"name": "beam", "id": 161, "trainId": 144},
+    {"name": "dishwasher, dish washer, dishwashing machine", "id": 747, "trainId": 145},
+    {"name": "plate", "id": 1919, "trainId": 146},
+    {"name": "screen, crt screen", "id": 3109, "trainId": 147},
+    {"name": "ruins", "id": 2179, "trainId": 148},
+    {"name": "washer, automatic washer, washing machine", "id": 2989, "trainId": 149},
+    {"name": "blanket, cover", "id": 206, "trainId": 150},
+    {"name": "plaything, toy", "id": 1930, "trainId": 151},
+    {"name": "food, solid food", "id": 1002, "trainId": 152},
+    {"name": "screen, silver screen, projection screen", "id": 2254, "trainId": 153},
+    {"name": "oven", "id": 1708, "trainId": 154},
+    {"name": "stage", "id": 2526, "trainId": 155},
+    {"name": "beacon, lighthouse, beacon light, pharos", "id": 160, "trainId": 156},
+    {"name": "umbrella", "id": 2901, "trainId": 157},
+    {"name": "sculpture", "id": 2262, "trainId": 158},
+    {"name": "aqueduct", "id": 44, "trainId": 159},
+    {"name": "container", "id": 597, "trainId": 160},
+    {"name": "scaffolding, staging", "id": 2235, "trainId": 161},
+    {"name": "hood, exhaust hood", "id": 1260, "trainId": 162},
+    {"name": "curb, curbing, kerb", "id": 682, "trainId": 163},
+    {"name": "roller coaster", "id": 2151, "trainId": 164},
+    {"name": "horse, equus caballus", "id": 3107, "trainId": 165},
+    {"name": "catwalk", "id": 432, "trainId": 166},
+    {"name": "glass, drinking glass", "id": 1098, "trainId": 167},
+    {"name": "vase", "id": 2932, "trainId": 168},
+    {"name": "central reservation", "id": 461, "trainId": 169},
+    {"name": "carousel", "id": 410, "trainId": 170},
+    {"name": "radiator", "id": 2046, "trainId": 171},
+    {"name": "closet", "id": 533, "trainId": 172},
+    {"name": "machine", "id": 1481, "trainId": 173},
+    {"name": "pier, wharf, wharfage, dock", "id": 1858, "trainId": 174},
+    {"name": "fan", "id": 894, "trainId": 175},
+    {"name": "inflatable bounce game", "id": 1322, "trainId": 176},
+    {"name": "pitch", "id": 1891, "trainId": 177},
+    {"name": "paper", "id": 1756, "trainId": 178},
+    {"name": "arcade, colonnade", "id": 49, "trainId": 179},
+    {"name": "hot tub", "id": 1272, "trainId": 180},
+    {"name": "helicopter", "id": 1229, "trainId": 181},
+    {"name": "tray", "id": 2850, "trainId": 182},
+    {"name": "partition, divider", "id": 1784, "trainId": 183},
+    {"name": "vineyard", "id": 2962, "trainId": 184},
+    {"name": "bowl", "id": 259, "trainId": 185},
+    {"name": "bullring", "id": 319, "trainId": 186},
+    {"name": "flag", "id": 954, "trainId": 187},
+    {"name": "pot", "id": 1974, "trainId": 188},
+    {"name": "footbridge, overcrossing, pedestrian bridge", "id": 1013, "trainId": 189},
+    {"name": "shower", "id": 2356, "trainId": 190},
+    {"name": "bag, traveling bag, travelling bag, grip, suitcase", "id": 97, "trainId": 191},
+    {"name": "bulletin board, notice board", "id": 318, "trainId": 192},
+    {"name": "confessional booth", "id": 592, "trainId": 193},
+    {"name": "trunk, tree trunk, bole", "id": 2885, "trainId": 194},
+    {"name": "forest", "id": 1017, "trainId": 195},
+    {"name": "elevator door", "id": 851, "trainId": 196},
+    {"name": "laptop, laptop computer", "id": 1407, "trainId": 197},
+    {"name": "instrument panel", "id": 1332, "trainId": 198},
+    {"name": "bucket, pail", "id": 303, "trainId": 199},
+    {"name": "tapestry, tapis", "id": 2714, "trainId": 200},
+    {"name": "platform", "id": 1924, "trainId": 201},
+    {"name": "jacket", "id": 1346, "trainId": 202},
+    {"name": "gate", "id": 1081, "trainId": 203},
+    {"name": "monitor, monitoring device", "id": 1583, "trainId": 204},
+    {
+        "name": "telephone booth, phone booth, call box, telephone box, telephone kiosk",
+        "id": 2727,
+        "trainId": 205,
+    },
+    {"name": "spotlight, spot", "id": 2509, "trainId": 206},
+    {"name": "ring", "id": 2123, "trainId": 207},
+    {"name": "control panel", "id": 602, "trainId": 208},
+    {"name": "blackboard, chalkboard", "id": 202, "trainId": 209},
+    {"name": "air conditioner, air conditioning", "id": 10, "trainId": 210},
+    {"name": "chest", "id": 490, "trainId": 211},
+    {"name": "clock", "id": 530, "trainId": 212},
+    {"name": "sand dune", "id": 2213, "trainId": 213},
+    {"name": "pipe, pipage, piping", "id": 1884, "trainId": 214},
+    {"name": "vault", "id": 2934, "trainId": 215},
+    {"name": "table football", "id": 2687, "trainId": 216},
+    {"name": "cannon", "id": 387, "trainId": 217},
+    {"name": "swimming pool, swimming bath, natatorium", "id": 2668, "trainId": 218},
+    {"name": "fluorescent, fluorescent fixture", "id": 982, "trainId": 219},
+    {"name": "statue", "id": 2547, "trainId": 220},
+    {
+        "name": "loudspeaker, speaker, speaker unit, loudspeaker system, speaker system",
+        "id": 1474,
+        "trainId": 221,
+    },
+    {"name": "exhibitor", "id": 877, "trainId": 222},
+    {"name": "ladder", "id": 1391, "trainId": 223},
+    {"name": "carport", "id": 414, "trainId": 224},
+    {"name": "dam", "id": 698, "trainId": 225},
+    {"name": "pulpit", "id": 2019, "trainId": 226},
+    {"name": "skylight, fanlight", "id": 2422, "trainId": 227},
+    {"name": "water tower", "id": 3010, "trainId": 228},
+    {"name": "grill, grille, grillwork", "id": 1139, "trainId": 229},
+    {"name": "display board", "id": 753, "trainId": 230},
+    {"name": "pane, pane of glass, window glass", "id": 1747, "trainId": 231},
+    {"name": "rubbish, trash, scrap", "id": 2175, "trainId": 232},
+    {"name": "ice rink", "id": 1301, "trainId": 233},
+    {"name": "fruit", "id": 1033, "trainId": 234},
+    {"name": "patio", "id": 1789, "trainId": 235},
+    {"name": "vending machine", "id": 2939, "trainId": 236},
+    {"name": "telephone, phone, telephone set", "id": 2730, "trainId": 237},
+    {"name": "net", "id": 1652, "trainId": 238},
+    {
+        "name": "backpack, back pack, knapsack, packsack, rucksack, haversack",
+        "id": 90,
+        "trainId": 239,
+    },
+    {"name": "jar", "id": 1349, "trainId": 240},
+    {"name": "track", "id": 2830, "trainId": 241},
+    {"name": "magazine", "id": 1485, "trainId": 242},
+    {"name": "shutter", "id": 2370, "trainId": 243},
+    {"name": "roof", "id": 2155, "trainId": 244},
+    {"name": "banner, streamer", "id": 118, "trainId": 245},
+    {"name": "landfill", "id": 1402, "trainId": 246},
+    {"name": "post", "id": 1957, "trainId": 247},
+    {"name": "altarpiece, reredos", "id": 3130, "trainId": 248},
+    {"name": "hat, chapeau, lid", "id": 1197, "trainId": 249},
+    {"name": "arch, archway", "id": 52, "trainId": 250},
+    {"name": "table game", "id": 2688, "trainId": 251},
+    {"name": "bag, handbag, pocketbook, purse", "id": 96, "trainId": 252},
+    {"name": "document, written document, papers", "id": 762, "trainId": 253},
+    {"name": "dome", "id": 772, "trainId": 254},
+    {"name": "pier", "id": 1857, "trainId": 255},
+    {"name": "shanties", "id": 2315, "trainId": 256},
+    {"name": "forecourt", "id": 1016, "trainId": 257},
+    {"name": "crane", "id": 643, "trainId": 258},
+    {"name": "dog, domestic dog, canis familiaris", "id": 3105, "trainId": 259},
+    {"name": "piano, pianoforte, forte-piano", "id": 1849, "trainId": 260},
+    {"name": "drawing", "id": 791, "trainId": 261},
+    {"name": "cabin", "id": 349, "trainId": 262},
+    {
+        "name": "ad, advertisement, advertizement, advertising, advertizing, advert",
+        "id": 6,
+        "trainId": 263,
+    },
+    {"name": "amphitheater, amphitheatre, coliseum", "id": 3114, "trainId": 264},
+    {"name": "monument", "id": 1587, "trainId": 265},
+    {"name": "henhouse", "id": 1233, "trainId": 266},
+    {"name": "cockpit", "id": 559, "trainId": 267},
+    {"name": "heater, warmer", "id": 1223, "trainId": 268},
+    {"name": "windmill, aerogenerator, wind generator", "id": 3049, "trainId": 269},
+    {"name": "pool", "id": 1943, "trainId": 270},
+    {"name": "elevator, lift", "id": 853, "trainId": 271},
+    {"name": "decoration, ornament, ornamentation", "id": 709, "trainId": 272},
+    {"name": "labyrinth", "id": 1390, "trainId": 273},
+    {"name": "text, textual matter", "id": 2748, "trainId": 274},
+    {"name": "printer", "id": 2007, "trainId": 275},
+    {"name": "mezzanine, first balcony", "id": 1546, "trainId": 276},
+    {"name": "mattress", "id": 1513, "trainId": 277},
+    {"name": "straw", "id": 2600, "trainId": 278},
+    {"name": "stalls", "id": 2538, "trainId": 279},
+    {"name": "patio, terrace", "id": 1790, "trainId": 280},
+    {"name": "billboard, hoarding", "id": 194, "trainId": 281},
+    {"name": "bus stop", "id": 326, "trainId": 282},
+    {"name": "trouser, pant", "id": 2877, "trainId": 283},
+    {"name": "console table, console", "id": 594, "trainId": 284},
+    {"name": "rack", "id": 2036, "trainId": 285},
+    {"name": "notebook", "id": 1662, "trainId": 286},
+    {"name": "shrine", "id": 2366, "trainId": 287},
+    {"name": "pantry", "id": 1754, "trainId": 288},
+    {"name": "cart", "id": 418, "trainId": 289},
+    {"name": "steam shovel", "id": 2553, "trainId": 290},
+    {"name": "porch", "id": 1951, "trainId": 291},
+    {"name": "postbox, mailbox, letter box", "id": 1963, "trainId": 292},
+    {"name": "figurine, statuette", "id": 918, "trainId": 293},
+    {"name": "recycling bin", "id": 2086, "trainId": 294},
+    {"name": "folding screen", "id": 997, "trainId": 295},
+    {"name": "telescope", "id": 2731, "trainId": 296},
+    {"name": "deck chair, beach chair", "id": 704, "trainId": 297},
+    {"name": "kennel", "id": 1365, "trainId": 298},
+    {"name": "coffee maker", "id": 569, "trainId": 299},
+    {"name": "altar, communion table, lord's table", "id": 3108, "trainId": 300},
+    {"name": "fish", "id": 948, "trainId": 301},
+    {"name": "easel", "id": 839, "trainId": 302},
+    {"name": "artificial golf green", "id": 63, "trainId": 303},
+    {"name": "iceberg", "id": 1305, "trainId": 304},
+    {"name": "candlestick, candle holder", "id": 378, "trainId": 305},
+    {"name": "shower stall, shower bath", "id": 2362, "trainId": 306},
+    {"name": "television stand", "id": 2734, "trainId": 307},
+    {
+        "name": "wall socket, wall plug, electric outlet, electrical outlet, outlet, electric receptacle",
+        "id": 2982,
+        "trainId": 308,
+    },
+    {"name": "skeleton", "id": 2398, "trainId": 309},
+    {"name": "grand piano, grand", "id": 1119, "trainId": 310},
+    {"name": "candy, confect", "id": 382, "trainId": 311},
+    {"name": "grille door", "id": 1141, "trainId": 312},
+    {"name": "pedestal, plinth, footstall", "id": 1805, "trainId": 313},
+    {"name": "jersey, t-shirt, tee shirt", "id": 3102, "trainId": 314},
+    {"name": "shoe", "id": 2341, "trainId": 315},
+    {"name": "gravestone, headstone, tombstone", "id": 1131, "trainId": 316},
+    {"name": "shanty", "id": 2316, "trainId": 317},
+    {"name": "structure", "id": 2626, "trainId": 318},
+    {"name": "rocking chair, rocker", "id": 3104, "trainId": 319},
+    {"name": "bird", "id": 198, "trainId": 320},
+    {"name": "place mat", "id": 1896, "trainId": 321},
+    {"name": "tomb", "id": 2800, "trainId": 322},
+    {"name": "big top", "id": 190, "trainId": 323},
+    {"name": "gas pump, gasoline pump, petrol pump, island dispenser", "id": 3131, "trainId": 324},
+    {"name": "lockers", "id": 1463, "trainId": 325},
+    {"name": "cage", "id": 357, "trainId": 326},
+    {"name": "finger", "id": 929, "trainId": 327},
+    {"name": "bleachers", "id": 209, "trainId": 328},
+    {"name": "ferris wheel", "id": 912, "trainId": 329},
+    {"name": "hairdresser chair", "id": 1164, "trainId": 330},
+    {"name": "mat", "id": 1509, "trainId": 331},
+    {"name": "stands", "id": 2539, "trainId": 332},
+    {"name": "aquarium, fish tank, marine museum", "id": 3116, "trainId": 333},
+    {"name": "streetcar, tram, tramcar, trolley, trolley car", "id": 2615, "trainId": 334},
+    {"name": "napkin, table napkin, serviette", "id": 1644, "trainId": 335},
+    {"name": "dummy", "id": 818, "trainId": 336},
+    {"name": "booklet, brochure, folder, leaflet, pamphlet", "id": 242, "trainId": 337},
+    {"name": "sand trap", "id": 2217, "trainId": 338},
+    {"name": "shop, store", "id": 2347, "trainId": 339},
+    {"name": "table cloth", "id": 2686, "trainId": 340},
+    {"name": "service station", "id": 2300, "trainId": 341},
+    {"name": "coffin", "id": 572, "trainId": 342},
+    {"name": "drawer", "id": 789, "trainId": 343},
+    {"name": "cages", "id": 358, "trainId": 344},
+    {"name": "slot machine, coin machine", "id": 2443, "trainId": 345},
+    {"name": "balcony", "id": 101, "trainId": 346},
+    {"name": "volleyball court", "id": 2969, "trainId": 347},
+    {"name": "table tennis", "id": 2692, "trainId": 348},
+    {"name": "control table", "id": 606, "trainId": 349},
+    {"name": "shirt", "id": 2339, "trainId": 350},
+    {"name": "merchandise, ware, product", "id": 1533, "trainId": 351},
+    {"name": "railway", "id": 2060, "trainId": 352},
+    {"name": "parterre", "id": 1782, "trainId": 353},
+    {"name": "chimney", "id": 495, "trainId": 354},
+    {"name": "can, tin, tin can", "id": 371, "trainId": 355},
+    {"name": "tanks", "id": 2707, "trainId": 356},
+    {"name": "fabric, cloth, material, textile", "id": 889, "trainId": 357},
+    {"name": "alga, algae", "id": 3156, "trainId": 358},
+    {"name": "system", "id": 2683, "trainId": 359},
+    {"name": "map", "id": 1499, "trainId": 360},
+    {"name": "greenhouse", "id": 1135, "trainId": 361},
+    {"name": "mug", "id": 1619, "trainId": 362},
+    {"name": "barbecue", "id": 125, "trainId": 363},
+    {"name": "trailer", "id": 2838, "trainId": 364},
+    {"name": "toilet tissue, toilet paper, bathroom tissue", "id": 2792, "trainId": 365},
+    {"name": "organ", "id": 1695, "trainId": 366},
+    {"name": "dishrag, dishcloth", "id": 746, "trainId": 367},
+    {"name": "island", "id": 1343, "trainId": 368},
+    {"name": "keyboard", "id": 1370, "trainId": 369},
+    {"name": "trench", "id": 2858, "trainId": 370},
+    {"name": "basket, basketball hoop, hoop", "id": 145, "trainId": 371},
+    {"name": "steering wheel, wheel", "id": 2565, "trainId": 372},
+    {"name": "pitcher, ewer", "id": 1892, "trainId": 373},
+    {"name": "goal", "id": 1103, "trainId": 374},
+    {"name": "bread, breadstuff, staff of life", "id": 286, "trainId": 375},
+    {"name": "beds", "id": 170, "trainId": 376},
+    {"name": "wood", "id": 3073, "trainId": 377},
+    {"name": "file cabinet", "id": 922, "trainId": 378},
+    {"name": "newspaper, paper", "id": 1655, "trainId": 379},
+    {"name": "motorboat", "id": 1602, "trainId": 380},
+    {"name": "rope", "id": 2160, "trainId": 381},
+    {"name": "guitar", "id": 1151, "trainId": 382},
+    {"name": "rubble", "id": 2176, "trainId": 383},
+    {"name": "scarf", "id": 2239, "trainId": 384},
+    {"name": "barrels", "id": 132, "trainId": 385},
+    {"name": "cap", "id": 394, "trainId": 386},
+    {"name": "leaves", "id": 1424, "trainId": 387},
+    {"name": "control tower", "id": 607, "trainId": 388},
+    {"name": "dashboard", "id": 700, "trainId": 389},
+    {"name": "bandstand", "id": 116, "trainId": 390},
+    {"name": "lectern", "id": 1425, "trainId": 391},
+    {"name": "switch, electric switch, electrical switch", "id": 2676, "trainId": 392},
+    {"name": "baseboard, mopboard, skirting board", "id": 141, "trainId": 393},
+    {"name": "shower room", "id": 2360, "trainId": 394},
+    {"name": "smoke", "id": 2449, "trainId": 395},
+    {"name": "faucet, spigot", "id": 897, "trainId": 396},
+    {"name": "bulldozer", "id": 317, "trainId": 397},
+    {"name": "saucepan", "id": 2228, "trainId": 398},
+    {"name": "shops", "id": 2351, "trainId": 399},
+    {"name": "meter", "id": 1543, "trainId": 400},
+    {"name": "crevasse", "id": 656, "trainId": 401},
+    {"name": "gear", "id": 1088, "trainId": 402},
+    {"name": "candelabrum, candelabra", "id": 373, "trainId": 403},
+    {"name": "sofa bed", "id": 2472, "trainId": 404},
+    {"name": "tunnel", "id": 2892, "trainId": 405},
+    {"name": "pallet", "id": 1740, "trainId": 406},
+    {"name": "wire, conducting wire", "id": 3067, "trainId": 407},
+    {"name": "kettle, boiler", "id": 1367, "trainId": 408},
+    {"name": "bidet", "id": 188, "trainId": 409},
+    {
+        "name": "baby buggy, baby carriage, carriage, perambulator, pram, stroller, go-cart, pushchair, pusher",
+        "id": 79,
+        "trainId": 410,
+    },
+    {"name": "music stand", "id": 1633, "trainId": 411},
+    {"name": "pipe, tube", "id": 1885, "trainId": 412},
+    {"name": "cup", "id": 677, "trainId": 413},
+    {"name": "parking meter", "id": 1779, "trainId": 414},
+    {"name": "ice hockey rink", "id": 1297, "trainId": 415},
+    {"name": "shelter", "id": 2334, "trainId": 416},
+    {"name": "weeds", "id": 3027, "trainId": 417},
+    {"name": "temple", "id": 2735, "trainId": 418},
+    {"name": "patty, cake", "id": 1791, "trainId": 419},
+    {"name": "ski slope", "id": 2405, "trainId": 420},
+    {"name": "panel", "id": 1748, "trainId": 421},
+    {"name": "wallet", "id": 2983, "trainId": 422},
+    {"name": "wheel", "id": 3035, "trainId": 423},
+    {"name": "towel rack, towel horse", "id": 2824, "trainId": 424},
+    {"name": "roundabout", "id": 2168, "trainId": 425},
+    {"name": "canister, cannister, tin", "id": 385, "trainId": 426},
+    {"name": "rod", "id": 2148, "trainId": 427},
+    {"name": "soap dispenser", "id": 2465, "trainId": 428},
+    {"name": "bell", "id": 175, "trainId": 429},
+    {"name": "canvas", "id": 390, "trainId": 430},
+    {"name": "box office, ticket office, ticket booth", "id": 268, "trainId": 431},
+    {"name": "teacup", "id": 2722, "trainId": 432},
+    {"name": "trellis", "id": 2857, "trainId": 433},
+    {"name": "workbench", "id": 3088, "trainId": 434},
+    {"name": "valley, vale", "id": 2926, "trainId": 435},
+    {"name": "toaster", "id": 2782, "trainId": 436},
+    {"name": "knife", "id": 1378, "trainId": 437},
+    {"name": "podium", "id": 1934, "trainId": 438},
+    {"name": "ramp", "id": 2072, "trainId": 439},
+    {"name": "tumble dryer", "id": 2889, "trainId": 440},
+    {"name": "fireplug, fire hydrant, plug", "id": 944, "trainId": 441},
+    {"name": "gym shoe, sneaker, tennis shoe", "id": 1158, "trainId": 442},
+    {"name": "lab bench", "id": 1383, "trainId": 443},
+    {"name": "equipment", "id": 867, "trainId": 444},
+    {"name": "rocky formation", "id": 2145, "trainId": 445},
+    {"name": "plastic", "id": 1915, "trainId": 446},
+    {"name": "calendar", "id": 361, "trainId": 447},
+    {"name": "caravan", "id": 402, "trainId": 448},
+    {"name": "check-in-desk", "id": 482, "trainId": 449},
+    {"name": "ticket counter", "id": 2761, "trainId": 450},
+    {"name": "brush", "id": 300, "trainId": 451},
+    {"name": "mill", "id": 1554, "trainId": 452},
+    {"name": "covered bridge", "id": 636, "trainId": 453},
+    {"name": "bowling alley", "id": 260, "trainId": 454},
+    {"name": "hanger", "id": 1186, "trainId": 455},
+    {"name": "excavator", "id": 871, "trainId": 456},
+    {"name": "trestle", "id": 2859, "trainId": 457},
+    {"name": "revolving door", "id": 2103, "trainId": 458},
+    {"name": "blast furnace", "id": 208, "trainId": 459},
+    {"name": "scale, weighing machine", "id": 2236, "trainId": 460},
+    {"name": "projector", "id": 2012, "trainId": 461},
+    {"name": "soap", "id": 2462, "trainId": 462},
+    {"name": "locker", "id": 1462, "trainId": 463},
+    {"name": "tractor", "id": 2832, "trainId": 464},
+    {"name": "stretcher", "id": 2617, "trainId": 465},
+    {"name": "frame", "id": 1024, "trainId": 466},
+    {"name": "grating", "id": 1129, "trainId": 467},
+    {"name": "alembic", "id": 18, "trainId": 468},
+    {"name": "candle, taper, wax light", "id": 376, "trainId": 469},
+    {"name": "barrier", "id": 134, "trainId": 470},
+    {"name": "cardboard", "id": 407, "trainId": 471},
+    {"name": "cave", "id": 434, "trainId": 472},
+    {"name": "puddle", "id": 2017, "trainId": 473},
+    {"name": "tarp", "id": 2717, "trainId": 474},
+    {"name": "price tag", "id": 2005, "trainId": 475},
+    {"name": "watchtower", "id": 2993, "trainId": 476},
+    {"name": "meters", "id": 1545, "trainId": 477},
+    {
+        "name": "light bulb, lightbulb, bulb, incandescent lamp, electric light, electric-light bulb",
+        "id": 1445,
+        "trainId": 478,
+    },
+    {"name": "tracks", "id": 2831, "trainId": 479},
+    {"name": "hair dryer", "id": 1161, "trainId": 480},
+    {"name": "skirt", "id": 2411, "trainId": 481},
+    {"name": "viaduct", "id": 2949, "trainId": 482},
+    {"name": "paper towel", "id": 1769, "trainId": 483},
+    {"name": "coat", "id": 552, "trainId": 484},
+    {"name": "sheet", "id": 2327, "trainId": 485},
+    {"name": "fire extinguisher, extinguisher, asphyxiator", "id": 939, "trainId": 486},
+    {"name": "water wheel", "id": 3013, "trainId": 487},
+    {"name": "pottery, clayware", "id": 1986, "trainId": 488},
+    {"name": "magazine rack", "id": 1486, "trainId": 489},
+    {"name": "teapot", "id": 2723, "trainId": 490},
+    {"name": "microphone, mike", "id": 1549, "trainId": 491},
+    {"name": "support", "id": 2649, "trainId": 492},
+    {"name": "forklift", "id": 1020, "trainId": 493},
+    {"name": "canyon", "id": 392, "trainId": 494},
+    {"name": "cash register, register", "id": 422, "trainId": 495},
+    {"name": "leaf, leafage, foliage", "id": 1419, "trainId": 496},
+    {"name": "remote control, remote", "id": 2099, "trainId": 497},
+    {"name": "soap dish", "id": 2464, "trainId": 498},
+    {"name": "windshield, windscreen", "id": 3058, "trainId": 499},
+    {"name": "cat", "id": 430, "trainId": 500},
+    {"name": "cue, cue stick, pool cue, pool stick", "id": 675, "trainId": 501},
+    {"name": "vent, venthole, vent-hole, blowhole", "id": 2941, "trainId": 502},
+    {"name": "videos", "id": 2955, "trainId": 503},
+    {"name": "shovel", "id": 2355, "trainId": 504},
+    {"name": "eaves", "id": 840, "trainId": 505},
+    {"name": "antenna, aerial, transmitting aerial", "id": 32, "trainId": 506},
+    {"name": "shipyard", "id": 2338, "trainId": 507},
+    {"name": "hen, biddy", "id": 1232, "trainId": 508},
+    {"name": "traffic cone", "id": 2834, "trainId": 509},
+    {"name": "washing machines", "id": 2991, "trainId": 510},
+    {"name": "truck crane", "id": 2879, "trainId": 511},
+    {"name": "cds", "id": 444, "trainId": 512},
+    {"name": "niche", "id": 1657, "trainId": 513},
+    {"name": "scoreboard", "id": 2246, "trainId": 514},
+    {"name": "briefcase", "id": 296, "trainId": 515},
+    {"name": "boot", "id": 245, "trainId": 516},
+    {"name": "sweater, jumper", "id": 2661, "trainId": 517},
+    {"name": "hay", "id": 1202, "trainId": 518},
+    {"name": "pack", "id": 1714, "trainId": 519},
+    {"name": "bottle rack", "id": 251, "trainId": 520},
+    {"name": "glacier", "id": 1095, "trainId": 521},
+    {"name": "pergola", "id": 1828, "trainId": 522},
+    {"name": "building materials", "id": 311, "trainId": 523},
+    {"name": "television camera", "id": 2732, "trainId": 524},
+    {"name": "first floor", "id": 947, "trainId": 525},
+    {"name": "rifle", "id": 2115, "trainId": 526},
+    {"name": "tennis table", "id": 2738, "trainId": 527},
+    {"name": "stadium", "id": 2525, "trainId": 528},
+    {"name": "safety belt", "id": 2194, "trainId": 529},
+    {"name": "cover", "id": 634, "trainId": 530},
+    {"name": "dish rack", "id": 740, "trainId": 531},
+    {"name": "synthesizer", "id": 2682, "trainId": 532},
+    {"name": "pumpkin", "id": 2020, "trainId": 533},
+    {"name": "gutter", "id": 1156, "trainId": 534},
+    {"name": "fruit stand", "id": 1036, "trainId": 535},
+    {"name": "ice floe, floe", "id": 1295, "trainId": 536},
+    {"name": "handle, grip, handgrip, hold", "id": 1181, "trainId": 537},
+    {"name": "wheelchair", "id": 3037, "trainId": 538},
+    {"name": "mousepad, mouse mat", "id": 1614, "trainId": 539},
+    {"name": "diploma", "id": 736, "trainId": 540},
+    {"name": "fairground ride", "id": 893, "trainId": 541},
+    {"name": "radio", "id": 2047, "trainId": 542},
+    {"name": "hotplate", "id": 1274, "trainId": 543},
+    {"name": "junk", "id": 1361, "trainId": 544},
+    {"name": "wheelbarrow", "id": 3036, "trainId": 545},
+    {"name": "stream", "id": 2606, "trainId": 546},
+    {"name": "toll plaza", "id": 2797, "trainId": 547},
+    {"name": "punching bag", "id": 2022, "trainId": 548},
+    {"name": "trough", "id": 2876, "trainId": 549},
+    {"name": "throne", "id": 2758, "trainId": 550},
+    {"name": "chair desk", "id": 472, "trainId": 551},
+    {"name": "weighbridge", "id": 3028, "trainId": 552},
+    {"name": "extractor fan", "id": 882, "trainId": 553},
+    {"name": "hanging clothes", "id": 1189, "trainId": 554},
+    {"name": "dish, dish aerial, dish antenna, saucer", "id": 743, "trainId": 555},
+    {"name": "alarm clock, alarm", "id": 3122, "trainId": 556},
+    {"name": "ski lift", "id": 2401, "trainId": 557},
+    {"name": "chain", "id": 468, "trainId": 558},
+    {"name": "garage", "id": 1061, "trainId": 559},
+    {"name": "mechanical shovel", "id": 1523, "trainId": 560},
+    {"name": "wine rack", "id": 3059, "trainId": 561},
+    {"name": "tramway", "id": 2843, "trainId": 562},
+    {"name": "treadmill", "id": 2853, "trainId": 563},
+    {"name": "menu", "id": 1529, "trainId": 564},
+    {"name": "block", "id": 214, "trainId": 565},
+    {"name": "well", "id": 3032, "trainId": 566},
+    {"name": "witness stand", "id": 3071, "trainId": 567},
+    {"name": "branch", "id": 277, "trainId": 568},
+    {"name": "duck", "id": 813, "trainId": 569},
+    {"name": "casserole", "id": 426, "trainId": 570},
+    {"name": "frying pan", "id": 1039, "trainId": 571},
+    {"name": "desk organizer", "id": 727, "trainId": 572},
+    {"name": "mast", "id": 1508, "trainId": 573},
+    {"name": "spectacles, specs, eyeglasses, glasses", "id": 2490, "trainId": 574},
+    {"name": "service elevator", "id": 2299, "trainId": 575},
+    {"name": "dollhouse", "id": 768, "trainId": 576},
+    {"name": "hammock", "id": 1172, "trainId": 577},
+    {"name": "clothes hanging", "id": 537, "trainId": 578},
+    {"name": "photocopier", "id": 1847, "trainId": 579},
+    {"name": "notepad", "id": 1664, "trainId": 580},
+    {"name": "golf cart", "id": 1110, "trainId": 581},
+    {"name": "footpath", "id": 1014, "trainId": 582},
+    {"name": "cross", "id": 662, "trainId": 583},
+    {"name": "baptismal font", "id": 121, "trainId": 584},
+    {"name": "boiler", "id": 227, "trainId": 585},
+    {"name": "skip", "id": 2410, "trainId": 586},
+    {"name": "rotisserie", "id": 2165, "trainId": 587},
+    {"name": "tables", "id": 2696, "trainId": 588},
+    {"name": "water mill", "id": 3005, "trainId": 589},
+    {"name": "helmet", "id": 1231, "trainId": 590},
+    {"name": "cover curtain", "id": 635, "trainId": 591},
+    {"name": "brick", "id": 292, "trainId": 592},
+    {"name": "table runner", "id": 2690, "trainId": 593},
+    {"name": "ashtray", "id": 65, "trainId": 594},
+    {"name": "street box", "id": 2607, "trainId": 595},
+    {"name": "stick", "id": 2574, "trainId": 596},
+    {"name": "hangers", "id": 1188, "trainId": 597},
+    {"name": "cells", "id": 456, "trainId": 598},
+    {"name": "urinal", "id": 2913, "trainId": 599},
+    {"name": "centerpiece", "id": 459, "trainId": 600},
+    {"name": "portable fridge", "id": 1955, "trainId": 601},
+    {"name": "dvds", "id": 827, "trainId": 602},
+    {"name": "golf club", "id": 1111, "trainId": 603},
+    {"name": "skirting board", "id": 2412, "trainId": 604},
+    {"name": "water cooler", "id": 2997, "trainId": 605},
+    {"name": "clipboard", "id": 528, "trainId": 606},
+    {"name": "camera, photographic camera", "id": 366, "trainId": 607},
+    {"name": "pigeonhole", "id": 1863, "trainId": 608},
+    {"name": "chips", "id": 500, "trainId": 609},
+    {"name": "food processor", "id": 1001, "trainId": 610},
+    {"name": "post box", "id": 1958, "trainId": 611},
+    {"name": "lid", "id": 1441, "trainId": 612},
+    {"name": "drum", "id": 809, "trainId": 613},
+    {"name": "blender", "id": 210, "trainId": 614},
+    {"name": "cave entrance", "id": 435, "trainId": 615},
+    {"name": "dental chair", "id": 718, "trainId": 616},
+    {"name": "obelisk", "id": 1674, "trainId": 617},
+    {"name": "canoe", "id": 388, "trainId": 618},
+    {"name": "mobile", "id": 1572, "trainId": 619},
+    {"name": "monitors", "id": 1584, "trainId": 620},
+    {"name": "pool ball", "id": 1944, "trainId": 621},
+    {"name": "cue rack", "id": 674, "trainId": 622},
+    {"name": "baggage carts", "id": 99, "trainId": 623},
+    {"name": "shore", "id": 2352, "trainId": 624},
+    {"name": "fork", "id": 1019, "trainId": 625},
+    {"name": "paper filer", "id": 1763, "trainId": 626},
+    {"name": "bicycle rack", "id": 185, "trainId": 627},
+    {"name": "coat rack", "id": 554, "trainId": 628},
+    {"name": "garland", "id": 1066, "trainId": 629},
+    {"name": "sports bag", "id": 2508, "trainId": 630},
+    {"name": "fish tank", "id": 951, "trainId": 631},
+    {"name": "towel dispenser", "id": 2822, "trainId": 632},
+    {"name": "carriage", "id": 415, "trainId": 633},
+    {"name": "brochure", "id": 297, "trainId": 634},
+    {"name": "plaque", "id": 1914, "trainId": 635},
+    {"name": "stringer", "id": 2619, "trainId": 636},
+    {"name": "iron", "id": 1338, "trainId": 637},
+    {"name": "spoon", "id": 2505, "trainId": 638},
+    {"name": "flag pole", "id": 955, "trainId": 639},
+    {"name": "toilet brush", "id": 2786, "trainId": 640},
+    {"name": "book stand", "id": 238, "trainId": 641},
+    {"name": "water faucet, water tap, tap, hydrant", "id": 3000, "trainId": 642},
+    {"name": "ticket office", "id": 2763, "trainId": 643},
+    {"name": "broom", "id": 299, "trainId": 644},
+    {"name": "dvd", "id": 822, "trainId": 645},
+    {"name": "ice bucket", "id": 1288, "trainId": 646},
+    {"name": "carapace, shell, cuticle, shield", "id": 3101, "trainId": 647},
+    {"name": "tureen", "id": 2894, "trainId": 648},
+    {"name": "folders", "id": 992, "trainId": 649},
+    {"name": "chess", "id": 489, "trainId": 650},
+    {"name": "root", "id": 2157, "trainId": 651},
+    {"name": "sewing machine", "id": 2309, "trainId": 652},
+    {"name": "model", "id": 1576, "trainId": 653},
+    {"name": "pen", "id": 1810, "trainId": 654},
+    {"name": "violin", "id": 2964, "trainId": 655},
+    {"name": "sweatshirt", "id": 2662, "trainId": 656},
+    {"name": "recycling materials", "id": 2087, "trainId": 657},
+    {"name": "mitten", "id": 1569, "trainId": 658},
+    {"name": "chopping board, cutting board", "id": 503, "trainId": 659},
+    {"name": "mask", "id": 1505, "trainId": 660},
+    {"name": "log", "id": 1468, "trainId": 661},
+    {"name": "mouse, computer mouse", "id": 1613, "trainId": 662},
+    {"name": "grill", "id": 1138, "trainId": 663},
+    {"name": "hole", "id": 1256, "trainId": 664},
+    {"name": "target", "id": 2715, "trainId": 665},
+    {"name": "trash bag", "id": 2846, "trainId": 666},
+    {"name": "chalk", "id": 477, "trainId": 667},
+    {"name": "sticks", "id": 2576, "trainId": 668},
+    {"name": "balloon", "id": 108, "trainId": 669},
+    {"name": "score", "id": 2245, "trainId": 670},
+    {"name": "hair spray", "id": 1162, "trainId": 671},
+    {"name": "roll", "id": 2149, "trainId": 672},
+    {"name": "runner", "id": 2183, "trainId": 673},
+    {"name": "engine", "id": 858, "trainId": 674},
+    {"name": "inflatable glove", "id": 1324, "trainId": 675},
+    {"name": "games", "id": 1055, "trainId": 676},
+    {"name": "pallets", "id": 1741, "trainId": 677},
+    {"name": "baskets", "id": 149, "trainId": 678},
+    {"name": "coop", "id": 615, "trainId": 679},
+    {"name": "dvd player", "id": 825, "trainId": 680},
+    {"name": "rocking horse", "id": 2143, "trainId": 681},
+    {"name": "buckets", "id": 304, "trainId": 682},
+    {"name": "bread rolls", "id": 283, "trainId": 683},
+    {"name": "shawl", "id": 2322, "trainId": 684},
+    {"name": "watering can", "id": 3017, "trainId": 685},
+    {"name": "spotlights", "id": 2510, "trainId": 686},
+    {"name": "post-it", "id": 1960, "trainId": 687},
+    {"name": "bowls", "id": 265, "trainId": 688},
+    {"name": "security camera", "id": 2282, "trainId": 689},
+    {"name": "runner cloth", "id": 2184, "trainId": 690},
+    {"name": "lock", "id": 1461, "trainId": 691},
+    {"name": "alarm, warning device, alarm system", "id": 3113, "trainId": 692},
+    {"name": "side", "id": 2372, "trainId": 693},
+    {"name": "roulette", "id": 2166, "trainId": 694},
+    {"name": "bone", "id": 232, "trainId": 695},
+    {"name": "cutlery", "id": 693, "trainId": 696},
+    {"name": "pool balls", "id": 1945, "trainId": 697},
+    {"name": "wheels", "id": 3039, "trainId": 698},
+    {"name": "spice rack", "id": 2494, "trainId": 699},
+    {"name": "plant pots", "id": 1908, "trainId": 700},
+    {"name": "towel ring", "id": 2827, "trainId": 701},
+    {"name": "bread box", "id": 280, "trainId": 702},
+    {"name": "video", "id": 2950, "trainId": 703},
+    {"name": "funfair", "id": 1044, "trainId": 704},
+    {"name": "breads", "id": 288, "trainId": 705},
+    {"name": "tripod", "id": 2863, "trainId": 706},
+    {"name": "ironing board", "id": 1342, "trainId": 707},
+    {"name": "skimmer", "id": 2409, "trainId": 708},
+    {"name": "hollow", "id": 1258, "trainId": 709},
+    {"name": "scratching post", "id": 2249, "trainId": 710},
+    {"name": "tricycle", "id": 2862, "trainId": 711},
+    {"name": "file box", "id": 920, "trainId": 712},
+    {"name": "mountain pass", "id": 1607, "trainId": 713},
+    {"name": "tombstones", "id": 2802, "trainId": 714},
+    {"name": "cooker", "id": 610, "trainId": 715},
+    {"name": "card game, cards", "id": 3129, "trainId": 716},
+    {"name": "golf bag", "id": 1108, "trainId": 717},
+    {"name": "towel paper", "id": 2823, "trainId": 718},
+    {"name": "chaise lounge", "id": 476, "trainId": 719},
+    {"name": "sun", "id": 2641, "trainId": 720},
+    {"name": "toilet paper holder", "id": 2788, "trainId": 721},
+    {"name": "rake", "id": 2070, "trainId": 722},
+    {"name": "key", "id": 1368, "trainId": 723},
+    {"name": "umbrella stand", "id": 2903, "trainId": 724},
+    {"name": "dartboard", "id": 699, "trainId": 725},
+    {"name": "transformer", "id": 2844, "trainId": 726},
+    {"name": "fireplace utensils", "id": 942, "trainId": 727},
+    {"name": "sweatshirts", "id": 2663, "trainId": 728},
+    {
+        "name": "cellular telephone, cellular phone, cellphone, cell, mobile phone",
+        "id": 457,
+        "trainId": 729,
+    },
+    {"name": "tallboy", "id": 2701, "trainId": 730},
+    {"name": "stapler", "id": 2540, "trainId": 731},
+    {"name": "sauna", "id": 2231, "trainId": 732},
+    {"name": "test tube", "id": 2746, "trainId": 733},
+    {"name": "palette", "id": 1738, "trainId": 734},
+    {"name": "shopping carts", "id": 2350, "trainId": 735},
+    {"name": "tools", "id": 2808, "trainId": 736},
+    {"name": "push button, push, button", "id": 2025, "trainId": 737},
+    {"name": "star", "id": 2541, "trainId": 738},
+    {"name": "roof rack", "id": 2156, "trainId": 739},
+    {"name": "barbed wire", "id": 126, "trainId": 740},
+    {"name": "spray", "id": 2512, "trainId": 741},
+    {"name": "ear", "id": 831, "trainId": 742},
+    {"name": "sponge", "id": 2503, "trainId": 743},
+    {"name": "racket", "id": 2039, "trainId": 744},
+    {"name": "tins", "id": 2774, "trainId": 745},
+    {"name": "eyeglasses", "id": 886, "trainId": 746},
+    {"name": "file", "id": 919, "trainId": 747},
+    {"name": "scarfs", "id": 2240, "trainId": 748},
+    {"name": "sugar bowl", "id": 2636, "trainId": 749},
+    {"name": "flip flop", "id": 963, "trainId": 750},
+    {"name": "headstones", "id": 1218, "trainId": 751},
+    {"name": "laptop bag", "id": 1406, "trainId": 752},
+    {"name": "leash", "id": 1420, "trainId": 753},
+    {"name": "climbing frame", "id": 526, "trainId": 754},
+    {"name": "suit hanger", "id": 2639, "trainId": 755},
+    {"name": "floor spotlight", "id": 975, "trainId": 756},
+    {"name": "plate rack", "id": 1921, "trainId": 757},
+    {"name": "sewer", "id": 2305, "trainId": 758},
+    {"name": "hard drive", "id": 1193, "trainId": 759},
+    {"name": "sprinkler", "id": 2517, "trainId": 760},
+    {"name": "tools box", "id": 2809, "trainId": 761},
+    {"name": "necklace", "id": 1647, "trainId": 762},
+    {"name": "bulbs", "id": 314, "trainId": 763},
+    {"name": "steel industry", "id": 2560, "trainId": 764},
+    {"name": "club", "id": 545, "trainId": 765},
+    {"name": "jack", "id": 1345, "trainId": 766},
+    {"name": "door bars", "id": 775, "trainId": 767},
+    {
+        "name": "control panel, instrument panel, control board, board, panel",
+        "id": 603,
+        "trainId": 768,
+    },
+    {"name": "hairbrush", "id": 1163, "trainId": 769},
+    {"name": "napkin holder", "id": 1641, "trainId": 770},
+    {"name": "office", "id": 1678, "trainId": 771},
+    {"name": "smoke detector", "id": 2450, "trainId": 772},
+    {"name": "utensils", "id": 2915, "trainId": 773},
+    {"name": "apron", "id": 42, "trainId": 774},
+    {"name": "scissors", "id": 2242, "trainId": 775},
+    {"name": "terminal", "id": 2741, "trainId": 776},
+    {"name": "grinder", "id": 1143, "trainId": 777},
+    {"name": "entry phone", "id": 862, "trainId": 778},
+    {"name": "newspaper stand", "id": 1654, "trainId": 779},
+    {"name": "pepper shaker", "id": 1826, "trainId": 780},
+    {"name": "onions", "id": 1689, "trainId": 781},
+    {
+        "name": "central processing unit, cpu, c p u , central processor, processor, mainframe",
+        "id": 3124,
+        "trainId": 782,
+    },
+    {"name": "tape", "id": 2710, "trainId": 783},
+    {"name": "bat", "id": 152, "trainId": 784},
+    {"name": "coaster", "id": 549, "trainId": 785},
+    {"name": "calculator", "id": 360, "trainId": 786},
+    {"name": "potatoes", "id": 1982, "trainId": 787},
+    {"name": "luggage rack", "id": 1478, "trainId": 788},
+    {"name": "salt", "id": 2203, "trainId": 789},
+    {"name": "street number", "id": 2612, "trainId": 790},
+    {"name": "viewpoint", "id": 2956, "trainId": 791},
+    {"name": "sword", "id": 2681, "trainId": 792},
+    {"name": "cd", "id": 437, "trainId": 793},
+    {"name": "rowing machine", "id": 2171, "trainId": 794},
+    {"name": "plug", "id": 1933, "trainId": 795},
+    {"name": "andiron, firedog, dog, dog-iron", "id": 3110, "trainId": 796},
+    {"name": "pepper", "id": 1824, "trainId": 797},
+    {"name": "tongs", "id": 2803, "trainId": 798},
+    {"name": "bonfire", "id": 234, "trainId": 799},
+    {"name": "dog dish", "id": 764, "trainId": 800},
+    {"name": "belt", "id": 177, "trainId": 801},
+    {"name": "dumbbells", "id": 817, "trainId": 802},
+    {"name": "videocassette recorder, vcr", "id": 3145, "trainId": 803},
+    {"name": "hook", "id": 1262, "trainId": 804},
+    {"name": "envelopes", "id": 864, "trainId": 805},
+    {"name": "shower faucet", "id": 2359, "trainId": 806},
+    {"name": "watch", "id": 2992, "trainId": 807},
+    {"name": "padlock", "id": 1725, "trainId": 808},
+    {"name": "swimming pool ladder", "id": 2667, "trainId": 809},
+    {"name": "spanners", "id": 2484, "trainId": 810},
+    {"name": "gravy boat", "id": 1133, "trainId": 811},
+    {"name": "notice board", "id": 1667, "trainId": 812},
+    {"name": "trash bags", "id": 2847, "trainId": 813},
+    {"name": "fire alarm", "id": 932, "trainId": 814},
+    {"name": "ladle", "id": 1392, "trainId": 815},
+    {"name": "stethoscope", "id": 2573, "trainId": 816},
+    {"name": "rocket", "id": 2140, "trainId": 817},
+    {"name": "funnel", "id": 1046, "trainId": 818},
+    {"name": "bowling pins", "id": 264, "trainId": 819},
+    {"name": "valve", "id": 2927, "trainId": 820},
+    {"name": "thermometer", "id": 2752, "trainId": 821},
+    {"name": "cups", "id": 679, "trainId": 822},
+    {"name": "spice jar", "id": 2493, "trainId": 823},
+    {"name": "night light", "id": 1658, "trainId": 824},
+    {"name": "soaps", "id": 2466, "trainId": 825},
+    {"name": "games table", "id": 1057, "trainId": 826},
+    {"name": "slotted spoon", "id": 2444, "trainId": 827},
+    {"name": "reel", "id": 2093, "trainId": 828},
+    {"name": "scourer", "id": 2248, "trainId": 829},
+    {"name": "sleeping robe", "id": 2432, "trainId": 830},
+    {"name": "desk mat", "id": 726, "trainId": 831},
+    {"name": "dumbbell", "id": 816, "trainId": 832},
+    {"name": "hammer", "id": 1171, "trainId": 833},
+    {"name": "tie", "id": 2766, "trainId": 834},
+    {"name": "typewriter", "id": 2900, "trainId": 835},
+    {"name": "shaker", "id": 2313, "trainId": 836},
+    {"name": "cheese dish", "id": 488, "trainId": 837},
+    {"name": "sea star", "id": 2265, "trainId": 838},
+    {"name": "racquet", "id": 2043, "trainId": 839},
+    {"name": "butane gas cylinder", "id": 332, "trainId": 840},
+    {"name": "paper weight", "id": 1771, "trainId": 841},
+    {"name": "shaving brush", "id": 2320, "trainId": 842},
+    {"name": "sunglasses", "id": 2646, "trainId": 843},
+    {"name": "gear shift", "id": 1089, "trainId": 844},
+    {"name": "towel rail", "id": 2826, "trainId": 845},
+    {"name": "adding machine, totalizer, totaliser", "id": 3148, "trainId": 846},
+]
+
+
+def loadAde20K(file):
+    fileseg = file.replace(".jpg", "_seg.png")
+    with Image.open(fileseg) as io:
+        seg = np.array(io)
+
+    R = seg[:, :, 0]
+    G = seg[:, :, 1]
+    ObjectClassMasks = (R / 10).astype(np.int32) * 256 + (G.astype(np.int32))
+
+    return {"img_name": file, "segm_name": fileseg, "class_mask": ObjectClassMasks}
+
+
+if __name__ == "__main__":
+    dataset_dir = Path(os.getenv("DETECTRON2_DATASETS", "datasets"))
+    index_file = dataset_dir / "ADE20K_2021_17_01" / "index_ade20k.pkl"
+    print('Caution: we only generate the validation set!')
+    with open(index_file, "rb") as f:
+        index_ade20k = pkl.load(f)
+
+    id_map = {}
+    for cat in ADE20K_SEM_SEG_FULL_CATEGORIES:
+        id_map[cat["id"]] = cat["trainId"]
+
+    # make output dir
+    for name in ["training", "validation"]:
+        image_dir = dataset_dir / "ADE20K_2021_17_01" / "images_detectron2" / name
+        image_dir.mkdir(parents=True, exist_ok=True)
+        annotation_dir = dataset_dir / "ADE20K_2021_17_01" / "annotations_detectron2" / name
+        annotation_dir.mkdir(parents=True, exist_ok=True)
+
+    # process image and gt
+    for i, (folder_name, file_name) in tqdm.tqdm(
+        enumerate(zip(index_ade20k["folder"], index_ade20k["filename"])),
+        total=len(index_ade20k["filename"]),
+    ):
+        split = "validation" if file_name.split("_")[1] == "val" else "training"
+        if split == 'training':
+            # FIXME: If you want to generate training set, delete this condition
+            continue
+        info = loadAde20K(str(dataset_dir / folder_name / file_name))
+
+        # resize image and label
+        img = np.asarray(Image.open(info["img_name"]))
+        lab = np.asarray(info["class_mask"])
+
+        h, w = img.shape[0], img.shape[1]
+        max_size = 512
+        resize = True
+        if w >= h > max_size:
+            h_new, w_new = max_size, round(w / float(h) * max_size)
+        elif h >= w > max_size:
+            h_new, w_new = round(h / float(w) * max_size), max_size
+        else:
+            resize = False
+
+        if resize:
+            img = cv2.resize(img, (w_new, h_new), interpolation=cv2.INTER_LINEAR)
+            lab = cv2.resize(lab, (w_new, h_new), interpolation=cv2.INTER_NEAREST)
+
+        assert img.dtype == np.uint8
+        assert lab.dtype == np.int32
+
+        # apply label conversion and save into uint16 images
+        output = np.zeros_like(lab, dtype=np.uint16) + 65535
+        for obj_id in np.unique(lab):
+            if obj_id in id_map:
+                output[lab == obj_id] = id_map[obj_id]
+
+        output_img = dataset_dir / "ADE20K_2021_17_01" / "images_detectron2" / split / file_name
+        output_lab = (
+            dataset_dir
+            / "ADE20K_2021_17_01"
+            / "annotations_detectron2"
+            / split
+            / file_name.replace(".jpg", ".tif")
+        )
+        Image.fromarray(img).save(output_img)
+
+        assert output.dtype == np.uint16
+        Image.fromarray(output).save(output_lab)

datasets/prepare_ade20k_sem_seg.py

@@ -0,0 +1,35 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import os
+from pathlib import Path
+
+import numpy as np
+import tqdm
+from PIL import Image
+
+
+def convert(input, output, index=None):
+    img = np.asarray(Image.open(input))
+    assert img.dtype == np.uint8
+    img = img - 1  # 0 (ignore) becomes 255. others are shifted by 1
+    if index is not None:
+        mapping = {i: k for k, i in enumerate(index)}
+        img = np.vectorize(lambda x: mapping[x] if x in mapping else 255)(
+            img.astype(np.float)
+        ).astype(np.uint8)
+    Image.fromarray(img).save(output)
+
+
+if __name__ == "__main__":
+    dataset_dir = (
+        Path(os.getenv("DETECTRON2_DATASETS", "datasets")) / "ADEChallengeData2016"
+    )
+    print('Caution: we only generate the validation set!')
+    for name in ["validation"]:
+        annotation_dir = dataset_dir / "annotations" / name
+        output_dir = dataset_dir / "annotations_detectron2" / name
+        output_dir.mkdir(parents=True, exist_ok=True)
+        for file in tqdm.tqdm(list(annotation_dir.iterdir())):
+            output_file = output_dir / file.name
+            convert(file, output_file)

datasets/prepare_coco_stuff_sem_seg.py

@@ -0,0 +1,219 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+# Modified by Feng Liang from
+# https://github.com/MendelXu/zsseg.baseline/blob/master/datasets/prepare_coco_stuff_164k_sem_seg.py
+
+import os
+import os.path as osp
+from pathlib import Path
+import tqdm
+from glob import glob
+
+import numpy as np
+from PIL import Image
+
+
+full_clsID_to_trID = {
+    0: 0,
+    1: 1,
+    2: 2,
+    3: 3,
+    4: 4,
+    5: 5,
+    6: 6,
+    7: 7,
+    8: 8,
+    9: 9,
+    10: 10,
+    12: 11,
+    13: 12,
+    14: 13,
+    15: 14,
+    16: 15,
+    17: 16,
+    18: 17,
+    19: 18,
+    20: 19,
+    21: 20,
+    22: 21,
+    23: 22,
+    24: 23,
+    26: 24,
+    27: 25,
+    30: 26,
+    31: 27,
+    32: 28,
+    33: 29,
+    34: 30,
+    35: 31,
+    36: 32,
+    37: 33,
+    38: 34,
+    39: 35,
+    40: 36,
+    41: 37,
+    42: 38,
+    43: 39,
+    45: 40,
+    46: 41,
+    47: 42,
+    48: 43,
+    49: 44,
+    50: 45,
+    51: 46,
+    52: 47,
+    53: 48,
+    54: 49,
+    55: 50,
+    56: 51,
+    57: 52,
+    58: 53,
+    59: 54,
+    60: 55,
+    61: 56,
+    62: 57,
+    63: 58,
+    64: 59,
+    66: 60,
+    69: 61,
+    71: 62,
+    72: 63,
+    73: 64,
+    74: 65,
+    75: 66,
+    76: 67,
+    77: 68,
+    78: 69,
+    79: 70,
+    80: 71,
+    81: 72,
+    83: 73,
+    84: 74,
+    85: 75,
+    86: 76,
+    87: 77,
+    88: 78,
+    89: 79,
+    91: 80,
+    92: 81,
+    93: 82,
+    94: 83,
+    95: 84,
+    96: 85,
+    97: 86,
+    98: 87,
+    99: 88,
+    100: 89,
+    101: 90,
+    102: 91,
+    103: 92,
+    104: 93,
+    105: 94,
+    106: 95,
+    107: 96,
+    108: 97,
+    109: 98,
+    110: 99,
+    111: 100,
+    112: 101,
+    113: 102,
+    114: 103,
+    115: 104,
+    116: 105,
+    117: 106,
+    118: 107,
+    119: 108,
+    120: 109,
+    121: 110,
+    122: 111,
+    123: 112,
+    124: 113,
+    125: 114,
+    126: 115,
+    127: 116,
+    128: 117,
+    129: 118,
+    130: 119,
+    131: 120,
+    132: 121,
+    133: 122,
+    134: 123,
+    135: 124,
+    136: 125,
+    137: 126,
+    138: 127,
+    139: 128,
+    140: 129,
+    141: 130,
+    142: 131,
+    143: 132,
+    144: 133,
+    145: 134,
+    146: 135,
+    147: 136,
+    148: 137,
+    149: 138,
+    150: 139,
+    151: 140,
+    152: 141,
+    153: 142,
+    154: 143,
+    155: 144,
+    156: 145,
+    157: 146,
+    158: 147,
+    159: 148,
+    160: 149,
+    161: 150,
+    162: 151,
+    163: 152,
+    164: 153,
+    165: 154,
+    166: 155,
+    167: 156,
+    168: 157,
+    169: 158,
+    170: 159,
+    171: 160,
+    172: 161,
+    173: 162,
+    174: 163,
+    175: 164,
+    176: 165,
+    177: 166,
+    178: 167,
+    179: 168,
+    180: 169,
+    181: 170,
+    255: 255,
+}
+
+def convert_to_trainID(
+    maskpath, out_mask_dir, is_train, clsID_to_trID=full_clsID_to_trID, suffix=""
+):
+    mask = np.array(Image.open(maskpath))
+    mask_copy = np.ones_like(mask, dtype=np.uint8) * 255
+    for clsID, trID in clsID_to_trID.items():
+        mask_copy[mask == clsID] = trID
+    seg_filename = (
+        osp.join(out_mask_dir, "train2017" + suffix, osp.basename(maskpath))
+        if is_train
+        else osp.join(out_mask_dir, "val2017" + suffix, osp.basename(maskpath))
+    )
+    if len(np.unique(mask_copy)) == 1 and np.unique(mask_copy)[0] == 255:
+        return
+    Image.fromarray(mask_copy).save(seg_filename, "PNG")
+
+
+
+if __name__ == "__main__":
+    dataset_dir = Path(os.getenv("DETECTRON2_DATASETS", "datasets"))
+    print('Caution: we only generate the training set!')
+    coco_path = dataset_dir / "coco"
+    mask_dir = coco_path / "stuffthingmaps"
+    out_mask_dir = coco_path / "stuffthingmaps_detectron2"
+    for name in ["train2017"]:
+        os.makedirs((out_mask_dir / name), exist_ok=True)
+        train_list = glob(osp.join(mask_dir, "train2017", "*.png"))
+        for file in tqdm.tqdm(train_list):
+            convert_to_trainID(file, out_mask_dir, is_train=True)

datasets/prepare_pascal_context.py

@@ -0,0 +1,69 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import tqdm
+import os
+import os.path as osp
+from pathlib import Path
+
+import numpy as np
+from PIL import Image
+import scipy.io
+
+def convert_pc59(mask_path, new_mask_path, pc59_dict):
+    mat = scipy.io.loadmat(mask_path)
+    mask = mat['LabelMap']
+
+    mask_copy = np.ones_like(mask, dtype=np.uint8) * 255
+    for trID, clsID in pc59_dict.items():
+        mask_copy[mask == clsID] = trID
+
+    min_value = np.amin(mask_copy)
+    assert min_value >= 0, print(min_value)
+    Image.fromarray(mask_copy).save(new_mask_path, "PNG")
+
+def convert_pc459(mask_path, new_mask_path):
+    mat = scipy.io.loadmat(mask_path)
+    mask = mat['LabelMap']
+    mask = mask - 1
+    min_value = np.amin(mask)
+    assert min_value >= 0, print(min_value)
+    Image.fromarray(mask).save(new_mask_path, "TIFF")
+
+
+if __name__ == "__main__":
+    dataset_dir = Path(os.getenv("DETECTRON2_DATASETS", "datasets"))
+    print('Caution: we only generate the validation set!')
+    pc_path = dataset_dir / "VOCdevkit/VOC2010"
+
+    val_list = open(pc_path / "pascalcontext_val.txt", "r")
+    pc459_labels = open(pc_path / "labels.txt", "r")
+    pc59_labels = open(pc_path / "59_labels.txt", "r")
+
+    pc459_dict = {}
+    for line in pc459_labels.readlines():
+        if ':' in line:
+            idx, name = line.split(':')
+            idx = int(idx.strip())
+            name = name.strip()
+            pc459_dict[name] = idx
+
+    pc59_dict = {}
+    for i, line in enumerate(pc59_labels.readlines()):
+        name = line.split(':')[-1].strip()
+        if name is not '':
+            pc59_dict[i] = pc459_dict[name]
+
+    pc459_dir = pc_path / "annotations_detectron2" / "pc459_val"
+    pc459_dir.mkdir(parents=True, exist_ok=True)
+    pc59_dir = pc_path / "annotations_detectron2" / "pc59_val"
+    pc59_dir.mkdir(parents=True, exist_ok=True)
+
+    for line in tqdm.tqdm(val_list.readlines()):
+        fileid = line.strip()
+        ori_mask = f'{pc_path}/trainval/{fileid}.mat'
+        pc459_dst = f'{pc459_dir}/{fileid}.tif'
+        pc59_dst = f'{pc59_dir}/{fileid}.png'
+        if osp.exists(ori_mask):
+            convert_pc459(ori_mask, pc459_dst)
+            convert_pc59(ori_mask, pc59_dst, pc59_dict)

datasets/prepare_voc_sem_seg.py

@@ -0,0 +1,71 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+# Modified by Feng Liang from https://github.com/MendelXu/zsseg.baseline/blob/master/datasets/prepare_voc_sem_seg.py
+
+import os
+import os.path as osp
+from pathlib import Path
+import tqdm
+
+import numpy as np
+from PIL import Image
+
+
+clsID_to_trID = {
+    0: 255,
+    1: 0,
+    2: 1,
+    3: 2,
+    4: 3,
+    5: 4,
+    6: 5,
+    7: 6,
+    8: 7,
+    9: 8,
+    10: 9,
+    11: 10,
+    12: 11,
+    13: 12,
+    14: 13,
+    15: 14,
+    16: 15,
+    17: 16,
+    18: 17,
+    19: 18,
+    20: 19,
+    255: 255,
+}
+
+def convert_to_trainID(
+    maskpath, out_mask_dir, is_train, clsID_to_trID=clsID_to_trID, suffix=""
+):
+    mask = np.array(Image.open(maskpath))
+    mask_copy = np.ones_like(mask, dtype=np.uint8) * 255
+    for clsID, trID in clsID_to_trID.items():
+        mask_copy[mask == clsID] = trID
+    seg_filename = (
+        osp.join(out_mask_dir, "train" + suffix, osp.basename(maskpath))
+        if is_train
+        else osp.join(out_mask_dir, "val" + suffix, osp.basename(maskpath))
+    )
+    if len(np.unique(mask_copy)) == 1 and np.unique(mask_copy)[0] == 255:
+        return
+    Image.fromarray(mask_copy).save(seg_filename, "PNG")
+
+
+
+if __name__ == "__main__":
+    dataset_dir = Path(os.getenv("DETECTRON2_DATASETS", "datasets"))
+    print('Caution: we only generate the validation set!')
+    voc_path = dataset_dir / "VOCdevkit" / "VOC2012"
+    out_mask_dir = voc_path / "annotations_detectron2"
+    out_image_dir = voc_path / "images_detectron2"
+    for name in ["val"]:
+        os.makedirs((out_mask_dir / name), exist_ok=True)
+        os.makedirs((out_image_dir / name), exist_ok=True)
+        val_list = [
+            osp.join(voc_path, "SegmentationClassAug", f + ".png")
+            for f in np.loadtxt(osp.join(voc_path, "ImageSets/Segmentation/val.txt"), dtype=np.str).tolist()
+        ]
+        for file in tqdm.tqdm(val_list):
+            convert_to_trainID(file, out_mask_dir, is_train=False)

demo.py

@@ -0,0 +1,123 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import argparse
+import glob
+import multiprocessing as mp
+import os
+import time
+import cv2
+import tqdm
+import numpy as np
+
+from detectron2.config import get_cfg
+
+from detectron2.projects.deeplab import add_deeplab_config
+from detectron2.data.detection_utils import read_image
+from detectron2.utils.logger import setup_logger
+from open_vocab_seg import add_ovseg_config
+
+from open_vocab_seg.utils import VisualizationDemo
+
+# constants
+WINDOW_NAME = "Open vocabulary segmentation"
+
+
+def setup_cfg(args):
+    # load config from file and command-line arguments
+    cfg = get_cfg()
+    # for poly lr schedule
+    add_deeplab_config(cfg)
+    add_ovseg_config(cfg)
+    cfg.merge_from_file(args.config_file)
+    cfg.merge_from_list(args.opts)
+    cfg.freeze()
+    return cfg
+
+
+def get_parser():
+    parser = argparse.ArgumentParser(description="Detectron2 demo for open vocabulary segmentation")
+    parser.add_argument(
+        "--config-file",
+        default="configs/ovseg_swinB_vitL_demo.yaml",
+        metavar="FILE",
+        help="path to config file",
+    )
+    parser.add_argument(
+        "--input",
+        nargs="+",
+        help="A list of space separated input images; "
+        "or a single glob pattern such as 'directory/*.jpg'",
+    )
+    parser.add_argument(
+        "--class-names",
+        nargs="+",
+        help="A list of user-defined class_names"
+    )
+    parser.add_argument(
+        "--output",
+        help="A file or directory to save output visualizations. "
+        "If not given, will show output in an OpenCV window.",
+    )
+    parser.add_argument(
+        "--opts",
+        help="Modify config options using the command-line 'KEY VALUE' pairs",
+        default=[],
+        nargs=argparse.REMAINDER,
+    )
+    return parser
+
+
+if __name__ == "__main__":
+    mp.set_start_method("spawn", force=True)
+    args = get_parser().parse_args()
+    setup_logger(name="fvcore")
+    logger = setup_logger()
+    logger.info("Arguments: " + str(args))
+
+    cfg = setup_cfg(args)
+
+    demo = VisualizationDemo(cfg)
+    class_names = args.class_names
+    if args.input:
+        if len(args.input) == 1:
+            args.input = glob.glob(os.path.expanduser(args.input[0]))
+            assert args.input, "The input path(s) was not found"
+        for path in tqdm.tqdm(args.input, disable=not args.output):
+            # use PIL, to be consistent with evaluation
+            start_time = time.time()
+            predictions, visualized_output_rgb, visualized_output_depth, visualized_output_rgb_sam, visualized_output_depth_sam = demo.run_on_image_sam(path, class_names)
+            logger.info(
+                "{}: {} in {:.2f}s".format(
+                    path,
+                    "detected {} instances".format(len(predictions["instances"]))
+                    if "instances" in predictions
+                    else "finished",
+                    time.time() - start_time,
+                )
+            )
+
+            if args.output:
+                if os.path.isdir(args.output):
+                    assert os.path.isdir(args.output), args.output
+                    out_filename = os.path.join(args.output, os.path.basename(path))
+                else:
+                    assert len(args.input) == 1, "Please specify a directory with args.output"
+                    out_filename = args.output
+                visualized_output_rgb.save('RGB_Semantic_SAM.png')
+                visualized_output_depth.save('Depth_Semantic_SAM.png')
+                visualized_output_rgb_sam.save('RGB_Semantic_SAM_Mask.png')
+                visualized_output_depth_sam.save('Depth_Semantic_SAM_Mask.png')
+                rgb_3d_sam = demo.get_xyzrgb('RGB_Semantic_SAM.png', path)
+                depth_3d_sam = demo.get_xyzrgb('Depth_Semantic_SAM.png', path)
+                rgb_3d_sam_mask = demo.get_xyzrgb('RGB_Semantic_SAM_Mask.png', path)
+                depth_3d_sam_mask = demo.get_xyzrgb('Depth_Semantic_SAM_Mask.png', path)
+                np.savez('xyzrgb.npz', rgb_3d_sam = rgb_3d_sam, depth_3d_sam = depth_3d_sam, rgb_3d_sam_mask = rgb_3d_sam_mask, depth_3d_sam_mask = depth_3d_sam_mask)
+                demo.render_3d_video('xyzrgb.npz', path)
+            else:
+                cv2.namedWindow(WINDOW_NAME, cv2.WINDOW_NORMAL)
+                cv2.imshow(WINDOW_NAME, visualized_output_rgb.get_image()[:, :, ::-1])
+                if cv2.waitKey(0) == 27:
+                    break  # esc to quit
+    else:
+        raise NotImplementedError

flagged/log.csv

@@ -0,0 +1,3 @@
+name,output,flag,username,timestamp
+t,/mnt/lustre/jkyang/PSG4D/segment_anything_sailvos3d/ov-seg/flagged/output/tmpii192qpn.png,,,2023-04-23 12:23:23.301078
+t,/mnt/lustre/jkyang/PSG4D/segment_anything_sailvos3d/ov-seg/flagged/output/tmpqm122tsi.png,,,2023-04-23 12:26:06.661559

open_vocab_seg/__init__.py

@@ -0,0 +1,9 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+from . import data
+from . import modeling
+from .config import add_ovseg_config
+
+from .test_time_augmentation import SemanticSegmentorWithTTA
+from .ovseg_model import OVSeg, OVSegDEMO

open_vocab_seg/config.py

@@ -0,0 +1,133 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+from detectron2.config import CfgNode as CN
+
+
+def add_mask_former_default_config(cfg):
+    # data config
+    # select the dataset mapper
+    cfg.INPUT.DATASET_MAPPER_NAME = "mask_former_semantic"
+    # Color augmentation
+    cfg.INPUT.COLOR_AUG_SSD = False
+    # We retry random cropping until no single category in semantic segmentation GT occupies more
+    # than `SINGLE_CATEGORY_MAX_AREA` part of the crop.
+    cfg.INPUT.CROP.SINGLE_CATEGORY_MAX_AREA = 1.0
+    # Pad image and segmentation GT in dataset mapper.
+    cfg.INPUT.SIZE_DIVISIBILITY = -1
+
+    # solver config
+    # test batch size
+    cfg.SOLVER.TEST_IMS_PER_BATCH = 1
+    # weight decay on embedding
+    cfg.SOLVER.WEIGHT_DECAY_EMBED = 0.0
+    # optimizer
+    cfg.SOLVER.OPTIMIZER = "ADAMW"
+    cfg.SOLVER.BACKBONE_MULTIPLIER = 0.1
+
+    # mask_former model config
+    cfg.MODEL.MASK_FORMER = CN()
+
+    # loss
+    cfg.MODEL.MASK_FORMER.DEEP_SUPERVISION = True
+    cfg.MODEL.MASK_FORMER.NO_OBJECT_WEIGHT = 0.1
+    cfg.MODEL.MASK_FORMER.DICE_WEIGHT = 1.0
+    cfg.MODEL.MASK_FORMER.MASK_WEIGHT = 20.0
+
+    # transformer config
+    cfg.MODEL.MASK_FORMER.NHEADS = 8
+    cfg.MODEL.MASK_FORMER.DROPOUT = 0.1
+    cfg.MODEL.MASK_FORMER.DIM_FEEDFORWARD = 2048
+    cfg.MODEL.MASK_FORMER.ENC_LAYERS = 0
+    cfg.MODEL.MASK_FORMER.DEC_LAYERS = 6
+    cfg.MODEL.MASK_FORMER.PRE_NORM = False
+
+    cfg.MODEL.MASK_FORMER.HIDDEN_DIM = 256
+    cfg.MODEL.MASK_FORMER.NUM_OBJECT_QUERIES = 100
+
+    cfg.MODEL.MASK_FORMER.TRANSFORMER_IN_FEATURE = "res5"
+    cfg.MODEL.MASK_FORMER.ENFORCE_INPUT_PROJ = False
+
+    # mask_former inference config
+    cfg.MODEL.MASK_FORMER.TEST = CN()
+    cfg.MODEL.MASK_FORMER.TEST.PANOPTIC_ON = False
+    cfg.MODEL.MASK_FORMER.TEST.OBJECT_MASK_THRESHOLD = 0.0
+    cfg.MODEL.MASK_FORMER.TEST.OVERLAP_THRESHOLD = 0.0
+    cfg.MODEL.MASK_FORMER.TEST.SEM_SEG_POSTPROCESSING_BEFORE_INFERENCE = False
+
+    # Sometimes `backbone.size_divisibility` is set to 0 for some backbone (e.g. ResNet)
+    # you can use this config to override
+    cfg.MODEL.MASK_FORMER.SIZE_DIVISIBILITY = 32
+
+    # pixel decoder config
+    cfg.MODEL.SEM_SEG_HEAD.MASK_DIM = 256
+    # adding transformer in pixel decoder
+    cfg.MODEL.SEM_SEG_HEAD.TRANSFORMER_ENC_LAYERS = 0
+    # pixel decoder
+    cfg.MODEL.SEM_SEG_HEAD.PIXEL_DECODER_NAME = "BasePixelDecoder"
+
+    # swin transformer backbone
+    cfg.MODEL.SWIN = CN()
+    cfg.MODEL.SWIN.PRETRAIN_IMG_SIZE = 224
+    cfg.MODEL.SWIN.PATCH_SIZE = 4
+    cfg.MODEL.SWIN.EMBED_DIM = 96
+    cfg.MODEL.SWIN.DEPTHS = [2, 2, 6, 2]
+    cfg.MODEL.SWIN.NUM_HEADS = [3, 6, 12, 24]
+    cfg.MODEL.SWIN.WINDOW_SIZE = 7
+    cfg.MODEL.SWIN.MLP_RATIO = 4.0
+    cfg.MODEL.SWIN.QKV_BIAS = True
+    cfg.MODEL.SWIN.QK_SCALE = None
+    cfg.MODEL.SWIN.NORM_INDICES = None
+    cfg.MODEL.SWIN.PROJECTION = False
+    cfg.MODEL.SWIN.PROJECT_DIM = 256
+    cfg.MODEL.SWIN.DROP_RATE = 0.0
+    cfg.MODEL.SWIN.ATTN_DROP_RATE = 0.0
+    cfg.MODEL.SWIN.DROP_PATH_RATE = 0.3
+    cfg.MODEL.SWIN.APE = False
+    cfg.MODEL.SWIN.PATCH_NORM = True
+    cfg.MODEL.SWIN.OUT_FEATURES = ["res2", "res3", "res4", "res5"]
+
+
+def add_our_config(cfg):
+    cfg.TEST.SLIDING_WINDOW = False
+    cfg.TEST.SLIDING_TILE_SIZE = 224
+    cfg.TEST.SLIDING_OVERLAP = 2 / 3.0
+    # whether to use dense crf
+    cfg.TEST.DENSE_CRF = False
+    cfg.DATASETS.SAMPLE_PER_CLASS = -1
+    cfg.DATASETS.SAMPLE_SEED = 0
+    # embedding head
+    cfg.MODEL.SEM_SEG_HEAD.EMBEDDING_DIM = 512
+    cfg.MODEL.SEM_SEG_HEAD.EMBED_HIDDEN_DIM = 1024
+    cfg.MODEL.SEM_SEG_HEAD.EMBED_LAYERS = 2
+    # clip_adapter
+    cfg.MODEL.CLIP_ADAPTER = CN()
+    cfg.MODEL.CLIP_ADAPTER.TEXT_TEMPLATES = "vild"
+    # for predefined
+    cfg.MODEL.CLIP_ADAPTER.PREDEFINED_PROMPT_TEMPLATES = ["a photo of a {}."]
+    # for learnable prompt
+    cfg.MODEL.CLIP_ADAPTER.PROMPT_CHECKPOINT = ""
+    cfg.MODEL.CLIP_ADAPTER.CLIP_MODEL_NAME = "ViT-B/16"
+    cfg.MODEL.CLIP_ADAPTER.MASK_FILL = "mean"
+    cfg.MODEL.CLIP_ADAPTER.MASK_EXPAND_RATIO = 1.0
+    cfg.MODEL.CLIP_ADAPTER.MASK_THR = 0.4
+    cfg.MODEL.CLIP_ADAPTER.MASK_MATTING = False
+    cfg.MODEL.CLIP_ADAPTER.REGION_RESIZED = True
+    cfg.MODEL.CLIP_ADAPTER.CLIP_ENSEMBLE = True
+    cfg.MODEL.CLIP_ADAPTER.CLIP_ENSEMBLE_WEIGHT = 0.7
+    # for mask prompt
+    cfg.MODEL.CLIP_ADAPTER.MASK_PROMPT_DEPTH = 3
+    cfg.MODEL.CLIP_ADAPTER.MASK_PROMPT_FWD = False
+
+    # wandb
+    cfg.WANDB = CN()
+    cfg.WANDB.PROJECT = "open_vocab_seg"
+    cfg.WANDB.NAME = None
+
+
+def add_ovseg_config(cfg):
+    """
+    Add config for open_vocab_seg.
+    """
+    add_mask_former_default_config(cfg)
+    add_our_config(cfg)

open_vocab_seg/data/__init__.py

@@ -0,0 +1,9 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+from .dataset_mappers import *
+from . import datasets
+from .build import (
+    build_detection_train_loader,
+    build_detection_test_loader,
+)

open_vocab_seg/data/augmentations.py

@@ -0,0 +1,202 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import math
+import numbers
+import numpy as np
+from detectron2.data.transforms.augmentation import Augmentation
+from detectron2.data.transforms.transform import (
+    CropTransform,
+    ResizeTransform,
+    TransformList,
+)
+from PIL import Image
+from fvcore.transforms.transform import PadTransform
+
+
+def mask2box(mask: np.ndarray):
+    # use naive way
+    row = np.nonzero(mask.sum(axis=0))[0]
+    if len(row) == 0:
+        return None
+    x1 = row.min()
+    x2 = row.max()
+    col = np.nonzero(mask.sum(axis=1))[0]
+    y1 = col.min()
+    y2 = col.max()
+    return x1, y1, x2 + 1 - x1, y2 + 1 - y1
+
+
+def expand_box(x, y, w, h, expand_ratio=1.0, max_h=None, max_w=None):
+    cx = x + 0.5 * w
+    cy = y + 0.5 * h
+    w = w * expand_ratio
+    h = h * expand_ratio
+    box = [cx - 0.5 * w, cy - 0.5 * h, cx + 0.5 * w, cy + 0.5 * h]
+    if max_h is not None:
+        box[1] = max(0, box[1])
+        box[3] = min(max_h - 1, box[3])
+    if max_w is not None:
+        box[0] = max(0, box[0])
+        box[2] = min(max_w - 1, box[2])
+    box[2] = box[2] - box[0]
+    box[3] = box[3] - box[1]
+
+    return [int(b) for b in box]
+
+
+class CropImageWithMask(Augmentation):
+    def __init__(self, expand_ratio=1.0, mode="choice"):
+        if isinstance(expand_ratio, numbers.Number):
+            expand_ratio = (expand_ratio, expand_ratio)
+        self.mode = mode
+        self.expand_ratio = expand_ratio
+        if self.mode == "range":
+            assert len(expand_ratio) == 2 and expand_ratio[0] < expand_ratio[1]
+
+    def get_transform(self, image, sem_seg, category_id):
+        input_size = image.shape[:2]
+        bin_mask = sem_seg == category_id
+        x, y, w, h = mask2box(bin_mask)
+        if self.mode == "choice":
+            expand_ratio = np.random.choice(self.expand_ratio)
+        else:
+            expand_ratio = np.random.uniform(self.expand_ratio[0], self.expand_ratio[1])
+        x, y, w, h = expand_box(x, y, w, h, expand_ratio, *input_size)
+        w = max(w, 1)
+        h = max(h, 1)
+        return CropTransform(x, y, w, h, input_size[1], input_size[0])
+
+
+class CropImageWithBox(Augmentation):
+    def __init__(self, expand_ratio=1.0, mode="choice"):
+        if isinstance(expand_ratio, numbers.Number):
+            expand_ratio = (expand_ratio, expand_ratio)
+        self.mode = mode
+        self.expand_ratio = expand_ratio
+        if self.mode == "range":
+            assert len(expand_ratio) == 2 and expand_ratio[0] < expand_ratio[1]
+
+    def get_transform(self, image, boxes):
+        input_size = image.shape[:2]
+        x, y, x2, y2 = boxes[0]
+        w = x2 - x + 1
+        h = y2 - y + 1
+        if self.mode == "choice":
+            expand_ratio = np.random.choice(self.expand_ratio)
+        else:
+            expand_ratio = np.random.uniform(self.expand_ratio[0], self.expand_ratio[1])
+        x, y, w, h = expand_box(x, y, w, h, expand_ratio, *input_size)
+        w = max(w, 1)
+        h = max(h, 1)
+        return CropTransform(x, y, w, h, input_size[1], input_size[0])
+
+
+class RandomResizedCrop(Augmentation):
+    def __init__(
+        self,
+        size,
+        scale=(0.08, 1.0),
+        ratio=(3.0 / 4.0, 4.0 / 3.0),
+        interpolation=Image.BILINEAR,
+    ):
+        if isinstance(size, int):
+            size = (size, size)
+        else:
+            assert isinstance(size, (tuple, list)) and len(size) == 2
+
+        self.size = size
+
+        self.scale = scale
+        self.ratio = ratio
+        self.interpolation = interpolation
+
+    def get_transform(self, image):
+        height, width = image.shape[:2]
+        area = height * width
+
+        log_ratio = np.log(np.array(self.ratio))
+        is_success = False
+        for _ in range(10):
+            target_area = area * np.random.uniform(self.scale[0], self.scale[1])
+            aspect_ratio = np.exp(np.random.uniform(log_ratio[0], log_ratio[1]))
+
+            w = int(round(math.sqrt(target_area * aspect_ratio)))
+            h = int(round(math.sqrt(target_area / aspect_ratio)))
+
+            if 0 < w <= width and 0 < h <= height:
+                i = np.random.randint(0, width - w + 1)
+                j = np.random.randint(0, height - h + 1)
+
+                is_success = True
+                break
+
+        if not is_success:
+            # Fallback to central crop
+            in_ratio = float(width) / float(height)
+            if in_ratio < min(self.ratio):
+                w = width
+                h = int(round(w / min(self.ratio)))
+            elif in_ratio > max(self.ratio):
+                h = height
+                w = int(round(h * max(self.ratio)))
+            else:  # whole image
+                w = width
+                h = height
+            i = (width - w) // 2
+            j = (height - h) // 2
+        return TransformList(
+            [
+                CropTransform(i, j, w, h, width, height),
+                ResizeTransform(
+                    h, w, self.size[1], self.size[0], interp=self.interpolation
+                ),
+            ]
+        )
+
+
+class CenterCrop(Augmentation):
+    def __init__(self, size, seg_ignore_label):
+        if isinstance(size, numbers.Number):
+            size = (int(size), int(size))
+        elif isinstance(size, (tuple, list)) and len(size) == 1:
+            size = (size[0], size[0])
+        self.size = size
+        self.seg_ignore_label = seg_ignore_label
+
+    def get_transform(self, image):
+
+        image_height, image_width = image.shape[:2]
+        crop_height, crop_width = self.size
+
+        transforms = []
+        if crop_width > image_width or crop_height > image_height:
+            padding_ltrb = [
+                (crop_width - image_width) // 2 if crop_width > image_width else 0,
+                (crop_height - image_height) // 2 if crop_height > image_height else 0,
+                (crop_width - image_width + 1) // 2 if crop_width > image_width else 0,
+                (crop_height - image_height + 1) // 2
+                if crop_height > image_height
+                else 0,
+            ]
+            transforms.append(
+                PadTransform(
+                    *padding_ltrb,
+                    orig_w=image_width,
+                    orig_h=image_height,
+                    seg_pad_value=self.seg_ignore_label
+                )
+            )
+            image_width, image_height = (
+                image_width + padding_ltrb[0] + padding_ltrb[2],
+                image_height + padding_ltrb[1] + padding_ltrb[3],
+            )
+
+        crop_top = int(round((image_height - crop_height) / 2.0))
+        crop_left = int(round((image_width - crop_width) / 2.0))
+        transforms.append(
+            CropTransform(
+                crop_left, crop_top, crop_width, crop_height, image_width, image_height
+            )
+        )
+        return TransformList(transforms)

open_vocab_seg/data/build.py

@@ -0,0 +1,344 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import itertools
+import logging
+import numpy as np
+from collections import Counter
+import torch.utils.data
+from tabulate import tabulate
+from termcolor import colored
+
+from detectron2.utils.logger import _log_api_usage, log_first_n
+from detectron2.data.catalog import DatasetCatalog, MetadataCatalog
+import torch.utils.data
+from detectron2.config import configurable
+from detectron2.data.build import (
+    build_batch_data_loader,
+    trivial_batch_collator,
+    load_proposals_into_dataset,
+    filter_images_with_only_crowd_annotations,
+    filter_images_with_few_keypoints,
+    print_instances_class_histogram,
+)
+
+from detectron2.data.common import DatasetFromList, MapDataset
+from detectron2.data.dataset_mapper import DatasetMapper
+from detectron2.data.detection_utils import check_metadata_consistency
+from detectron2.data.samplers import (
+    InferenceSampler,
+    RandomSubsetTrainingSampler,
+    RepeatFactorTrainingSampler,
+    TrainingSampler,
+)
+
+"""
+This file contains the default logic to build a dataloader for training or testing.
+"""
+
+__all__ = [
+    "build_detection_train_loader",
+    "build_detection_test_loader",
+]
+
+
+def print_classification_instances_class_histogram(dataset_dicts, class_names):
+    """
+    Args:
+        dataset_dicts (list[dict]): list of dataset dicts.
+        class_names (list[str]): list of class names (zero-indexed).
+    """
+    num_classes = len(class_names)
+    hist_bins = np.arange(num_classes + 1)
+    histogram = np.zeros((num_classes,), dtype=np.int)
+    for entry in dataset_dicts:
+        classes = np.asarray([entry["category_id"]], dtype=np.int)
+        if len(classes):
+            assert classes.min() >= 0, f"Got an invalid category_id={classes.min()}"
+            assert (
+                classes.max() < num_classes
+            ), f"Got an invalid category_id={classes.max()} for a dataset of {num_classes} classes"
+        histogram += np.histogram(classes, bins=hist_bins)[0]
+
+    N_COLS = min(6, len(class_names) * 2)
+
+    def short_name(x):
+        # make long class names shorter. useful for lvis
+        if len(x) > 13:
+            return x[:11] + ".."
+        return x
+
+    data = list(
+        itertools.chain(
+            *[[short_name(class_names[i]), int(v)] for i, v in enumerate(histogram)]
+        )
+    )
+    total_num_instances = sum(data[1::2])
+    data.extend([None] * (N_COLS - (len(data) % N_COLS)))
+    if num_classes > 1:
+        data.extend(["total", total_num_instances])
+    data = itertools.zip_longest(*[data[i::N_COLS] for i in range(N_COLS)])
+    table = tabulate(
+        data,
+        headers=["category", "#instances"] * (N_COLS // 2),
+        tablefmt="pipe",
+        numalign="left",
+        stralign="center",
+    )
+    log_first_n(
+        logging.INFO,
+        "Distribution of instances among all {} categories:\n".format(num_classes)
+        + colored(table, "cyan"),
+        key="message",
+    )
+
+
+def wrap_metas(dataset_dict, **kwargs):
+    def _assign_attr(data_dict: dict, **kwargs):
+        assert not any(
+            [key in data_dict for key in kwargs]
+        ), "Assigned attributes should not exist in the original sample."
+        data_dict.update(kwargs)
+        return data_dict
+
+    return [_assign_attr(sample, meta=kwargs) for sample in dataset_dict]
+
+
+def get_detection_dataset_dicts(
+    names, filter_empty=True, min_keypoints=0, proposal_files=None
+):
+    """
+    Load and prepare dataset dicts for instance detection/segmentation and semantic segmentation.
+
+    Args:
+        names (str or list[str]): a dataset name or a list of dataset names
+        filter_empty (bool): whether to filter out images without instance annotations
+        min_keypoints (int): filter out images with fewer keypoints than
+            `min_keypoints`. Set to 0 to do nothing.
+        proposal_files (list[str]): if given, a list of object proposal files
+            that match each dataset in `names`.
+
+    Returns:
+        list[dict]: a list of dicts following the standard dataset dict format.
+    """
+    if isinstance(names, str):
+        names = [names]
+    assert len(names), names
+    dataset_dicts = [
+        wrap_metas(DatasetCatalog.get(dataset_name), dataset_name=dataset_name)
+        for dataset_name in names
+    ]
+    for dataset_name, dicts in zip(names, dataset_dicts):
+        assert len(dicts), "Dataset '{}' is empty!".format(dataset_name)
+
+    if proposal_files is not None:
+        assert len(names) == len(proposal_files)
+        # load precomputed proposals from proposal files
+        dataset_dicts = [
+            load_proposals_into_dataset(dataset_i_dicts, proposal_file)
+            for dataset_i_dicts, proposal_file in zip(dataset_dicts, proposal_files)
+        ]
+
+    dataset_dicts = list(itertools.chain.from_iterable(dataset_dicts))
+
+    has_instances = "annotations" in dataset_dicts[0]
+    if filter_empty and has_instances:
+        dataset_dicts = filter_images_with_only_crowd_annotations(dataset_dicts)
+    if min_keypoints > 0 and has_instances:
+        dataset_dicts = filter_images_with_few_keypoints(dataset_dicts, min_keypoints)
+
+    if has_instances:
+        try:
+            class_names = MetadataCatalog.get(names[0]).thing_classes
+            check_metadata_consistency("thing_classes", names)
+            print_instances_class_histogram(dataset_dicts, class_names)
+        except AttributeError:  # class names are not available for this dataset
+            pass
+
+    assert len(dataset_dicts), "No valid data found in {}.".format(",".join(names))
+    return dataset_dicts
+
+
+def _train_loader_from_config(cfg, mapper=None, *, dataset=None, sampler=None):
+    if dataset is None:
+        dataset = get_detection_dataset_dicts(
+            cfg.DATASETS.TRAIN,
+            filter_empty=cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS,
+            min_keypoints=cfg.MODEL.ROI_KEYPOINT_HEAD.MIN_KEYPOINTS_PER_IMAGE
+            if cfg.MODEL.KEYPOINT_ON
+            else 0,
+            proposal_files=cfg.DATASETS.PROPOSAL_FILES_TRAIN
+            if cfg.MODEL.LOAD_PROPOSALS
+            else None,
+        )
+        _log_api_usage("dataset." + cfg.DATASETS.TRAIN[0])
+
+    if mapper is None:
+        mapper = DatasetMapper(cfg, True)
+
+    if sampler is None:
+        sampler_name = cfg.DATALOADER.SAMPLER_TRAIN
+        logger = logging.getLogger(__name__)
+        logger.info("Using training sampler {}".format(sampler_name))
+        if sampler_name == "TrainingSampler":
+            sampler = TrainingSampler(len(dataset))
+        elif sampler_name == "RepeatFactorTrainingSampler":
+            repeat_factors = (
+                RepeatFactorTrainingSampler.repeat_factors_from_category_frequency(
+                    dataset, cfg.DATALOADER.REPEAT_THRESHOLD
+                )
+            )
+            sampler = RepeatFactorTrainingSampler(repeat_factors)
+        elif sampler_name == "RandomSubsetTrainingSampler":
+            sampler = RandomSubsetTrainingSampler(
+                len(dataset), cfg.DATALOADER.RANDOM_SUBSET_RATIO
+            )
+        else:
+            raise ValueError("Unknown training sampler: {}".format(sampler_name))
+
+    return {
+        "dataset": dataset,
+        "sampler": sampler,
+        "mapper": mapper,
+        "total_batch_size": cfg.SOLVER.IMS_PER_BATCH,
+        "aspect_ratio_grouping": cfg.DATALOADER.ASPECT_RATIO_GROUPING,
+        "num_workers": cfg.DATALOADER.NUM_WORKERS,
+    }
+
+
+# TODO can allow dataset as an iterable or IterableDataset to make this function more general
+@configurable(from_config=_train_loader_from_config)
+def build_detection_train_loader(
+    dataset,
+    *,
+    mapper,
+    sampler=None,
+    total_batch_size,
+    aspect_ratio_grouping=True,
+    num_workers=0,
+):
+    """
+    Build a dataloader for object detection with some default features.
+    This interface is experimental.
+
+    Args:
+        dataset (list or torch.utils.data.Dataset): a list of dataset dicts,
+            or a map-style pytorch dataset. They can be obtained by using
+            :func:`DatasetCatalog.get` or :func:`get_detection_dataset_dicts`.
+        mapper (callable): a callable which takes a sample (dict) from dataset and
+            returns the format to be consumed by the model.
+            When using cfg, the default choice is ``DatasetMapper(cfg, is_train=True)``.
+        sampler (torch.utils.data.sampler.Sampler or None): a sampler that produces
+            indices to be applied on ``dataset``. Default to :class:`TrainingSampler`,
+            which coordinates an infinite random shuffle sequence across all workers.
+        total_batch_size (int): total batch size across all workers. Batching
+            simply puts data into a list.
+        aspect_ratio_grouping (bool): whether to group images with similar
+            aspect ratio for efficiency. When enabled, it requires each
+            element in dataset be a dict with keys "width" and "height".
+        num_workers (int): number of parallel data loading workers
+
+    Returns:
+        torch.utils.data.DataLoader:
+            a dataloader. Each output from it is a ``list[mapped_element]`` of length
+            ``total_batch_size / num_workers``, where ``mapped_element`` is produced
+            by the ``mapper``.
+    """
+    if isinstance(dataset, list):
+        dataset = DatasetFromList(dataset, copy=False)
+    if mapper is not None:
+        dataset = MapDataset(dataset, mapper)
+    if sampler is None:
+        sampler = TrainingSampler(len(dataset))
+    assert isinstance(sampler, torch.utils.data.sampler.Sampler)
+    return build_batch_data_loader(
+        dataset,
+        sampler,
+        total_batch_size,
+        aspect_ratio_grouping=aspect_ratio_grouping,
+        num_workers=num_workers,
+    )
+
+
+def _test_loader_from_config(cfg, dataset_name, mapper=None):
+    """
+    Uses the given `dataset_name` argument (instead of the names in cfg), because the
+    standard practice is to evaluate each test set individually (not combining them).
+    """
+    if isinstance(dataset_name, str):
+        dataset_name = [dataset_name]
+
+    dataset = get_detection_dataset_dicts(
+        dataset_name,
+        filter_empty=False,
+        proposal_files=[
+            cfg.DATASETS.PROPOSAL_FILES_TEST[list(cfg.DATASETS.TEST).index(x)]
+            for x in dataset_name
+        ]
+        if cfg.MODEL.LOAD_PROPOSALS
+        else None,
+    )
+    if mapper is None:
+        mapper = DatasetMapper(cfg, False)
+    return {
+        "dataset": dataset,
+        "mapper": mapper,
+        "num_workers": 0,
+        "samples_per_gpu": cfg.SOLVER.TEST_IMS_PER_BATCH,
+    }
+
+
+@configurable(from_config=_test_loader_from_config)
+def build_detection_test_loader(
+    dataset, *, mapper, sampler=None, num_workers=0, samples_per_gpu=1
+):
+    """
+    Similar to `build_detection_train_loader`, but uses a batch size of 1,
+    and :class:`InferenceSampler`. This sampler coordinates all workers to
+    produce the exact set of all samples.
+    This interface is experimental.
+
+    Args:
+        dataset (list or torch.utils.data.Dataset): a list of dataset dicts,
+            or a map-style pytorch dataset. They can be obtained by using
+            :func:`DatasetCatalog.get` or :func:`get_detection_dataset_dicts`.
+        mapper (callable): a callable which takes a sample (dict) from dataset
+           and returns the format to be consumed by the model.
+           When using cfg, the default choice is ``DatasetMapper(cfg, is_train=False)``.
+        sampler (torch.utils.data.sampler.Sampler or None): a sampler that produces
+            indices to be applied on ``dataset``. Default to :class:`InferenceSampler`,
+            which splits the dataset across all workers.
+        num_workers (int): number of parallel data loading workers
+
+    Returns:
+        DataLoader: a torch DataLoader, that loads the given detection
+        dataset, with test-time transformation and batching.
+
+    Examples:
+    ::
+        data_loader = build_detection_test_loader(
+            DatasetRegistry.get("my_test"),
+            mapper=DatasetMapper(...))
+
+        # or, instantiate with a CfgNode:
+        data_loader = build_detection_test_loader(cfg, "my_test")
+    """
+    if isinstance(dataset, list):
+        dataset = DatasetFromList(dataset, copy=False)
+    if mapper is not None:
+        dataset = MapDataset(dataset, mapper)
+    if sampler is None:
+        sampler = InferenceSampler(len(dataset))
+    # Always use 1 image per worker during inference since this is the
+    # standard when reporting inference time in papers.
+    batch_sampler = torch.utils.data.sampler.BatchSampler(
+        sampler, samples_per_gpu, drop_last=False
+    )
+    data_loader = torch.utils.data.DataLoader(
+        dataset,
+        num_workers=num_workers,
+        batch_sampler=batch_sampler,
+        collate_fn=trivial_batch_collator,
+    )
+    return data_loader
+

open_vocab_seg/data/dataset_mappers/__init__.py

@@ -0,0 +1,4 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+from .mask_former_semantic_dataset_mapper import MaskFormerSemanticDatasetMapper

open_vocab_seg/data/dataset_mappers/mask_former_semantic_dataset_mapper.py

@@ -0,0 +1,208 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+# Copyright (c) Meta Platforms, Inc. All Rights Reserved
+
+import copy
+import logging
+
+import numpy as np
+import torch
+from torch.nn import functional as F
+
+from detectron2.config import configurable
+from detectron2.data import MetadataCatalog
+from detectron2.data import detection_utils as utils
+from detectron2.data import transforms as T
+from detectron2.projects.point_rend import ColorAugSSDTransform
+from detectron2.structures import BitMasks, Instances
+
+__all__ = ["MaskFormerSemanticDatasetMapper"]
+
+
+class MaskFormerSemanticDatasetMapper:
+    """
+    A callable which takes a dataset dict in Detectron2 Dataset format,
+    and map it into a format used by MaskFormer for semantic segmentation.
+
+    The callable currently does the following:
+
+    1. Read the image from "file_name"
+    2. Applies geometric transforms to the image and annotation
+    3. Find and applies suitable cropping to the image and annotation
+    4. Prepare image and annotation to Tensors
+    """
+
+    @configurable
+    def __init__(
+        self,
+        is_train=True,
+        *,
+        augmentations,
+        image_format,
+        ignore_label,
+        size_divisibility,
+    ):
+        """
+        NOTE: this interface is experimental.
+        Args:
+            is_train: for training or inference
+            augmentations: a list of augmentations or deterministic transforms to apply
+            image_format: an image format supported by :func:`detection_utils.read_image`.
+            ignore_label: the label that is ignored to evaluation
+            size_divisibility: pad image size to be divisible by this value
+        """
+        self.is_train = is_train
+        self.tfm_gens = augmentations
+        self.img_format = image_format
+        self.ignore_label = ignore_label
+        self.size_divisibility = size_divisibility
+
+        logger = logging.getLogger(__name__)
+        mode = "training" if is_train else "inference"
+        logger.info(
+            f"[{self.__class__.__name__}] Augmentations used in {mode}: {augmentations}"
+        )
+
+    @classmethod
+    def from_config(cls, cfg, is_train=True):
+        # Build augmentation
+        if is_train:
+            augs = [
+                T.ResizeShortestEdge(
+                    cfg.INPUT.MIN_SIZE_TRAIN,
+                    cfg.INPUT.MAX_SIZE_TRAIN,
+                    cfg.INPUT.MIN_SIZE_TRAIN_SAMPLING,
+                )
+            ]
+            if cfg.INPUT.CROP.ENABLED:
+                augs.append(
+                    T.RandomCrop_CategoryAreaConstraint(
+                        cfg.INPUT.CROP.TYPE,
+                        cfg.INPUT.CROP.SIZE,
+                        cfg.INPUT.CROP.SINGLE_CATEGORY_MAX_AREA,
+                        cfg.MODEL.SEM_SEG_HEAD.IGNORE_VALUE,
+                    )
+                )
+            if cfg.INPUT.COLOR_AUG_SSD:
+                augs.append(ColorAugSSDTransform(img_format=cfg.INPUT.FORMAT))
+            augs.append(T.RandomFlip())
+
+            # Assume always applies to the training set.
+            dataset_names = cfg.DATASETS.TRAIN
+        else:
+            min_size = cfg.INPUT.MIN_SIZE_TEST
+            max_size = cfg.INPUT.MAX_SIZE_TEST
+            sample_style = "choice"
+            augs = [T.ResizeShortestEdge(min_size, max_size, sample_style)]
+            dataset_names = cfg.DATASETS.TEST
+        meta = MetadataCatalog.get(dataset_names[0])
+        ignore_label = meta.ignore_label
+
+        ret = {
+            "is_train": is_train,
+            "augmentations": augs,
+            "image_format": cfg.INPUT.FORMAT,
+            "ignore_label": ignore_label,
+            "size_divisibility": cfg.INPUT.SIZE_DIVISIBILITY if is_train else -1,
+        }
+        return ret
+
+    def __call__(self, dataset_dict):
+        """
+        Args:
+            dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
+
+        Returns:
+            dict: a format that builtin models in detectron2 accept
+        """
+        # assert self.is_train, "MaskFormerSemanticDatasetMapper should only be used for training!"
+
+        dataset_dict = copy.deepcopy(dataset_dict)  # it will be modified by code below
+        image = utils.read_image(dataset_dict["file_name"], format=self.img_format)
+        utils.check_image_size(dataset_dict, image)
+
+        if "sem_seg_file_name" in dataset_dict:
+            # PyTorch transformation not implemented for uint16, so converting it to double first
+            sem_seg_gt = utils.read_image(dataset_dict.pop("sem_seg_file_name")).astype(
+                "double"
+            )
+        else:
+            sem_seg_gt = None
+
+        if sem_seg_gt is None:
+            raise ValueError(
+                "Cannot find 'sem_seg_file_name' for semantic segmentation dataset {}.".format(
+                    dataset_dict["file_name"]
+                )
+            )
+
+        aug_input = T.AugInput(image, sem_seg=sem_seg_gt)
+        aug_input, transforms = T.apply_transform_gens(self.tfm_gens, aug_input)
+        image = aug_input.image
+        sem_seg_gt = aug_input.sem_seg
+
+        # Pad image and segmentation label here!
+        image = torch.as_tensor(np.ascontiguousarray(image.transpose(2, 0, 1)))
+        if sem_seg_gt is not None:
+            sem_seg_gt = torch.as_tensor(sem_seg_gt.astype("long"))
+
+        if self.size_divisibility > 0:
+            image_size = (image.shape[-2], image.shape[-1])
+            padding_size = [
+                0,
+                self.size_divisibility - image_size[1],
+                0,
+                self.size_divisibility - image_size[0],
+            ]
+            image = F.pad(image, padding_size, value=128).contiguous()
+            if sem_seg_gt is not None:
+                sem_seg_gt = F.pad(
+                    sem_seg_gt, padding_size, value=self.ignore_label
+                ).contiguous()
+
+        image_shape = (image.shape[-2], image.shape[-1])  # h, w
+
+        # Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
+        # but not efficient on large generic data structures due to the use of pickle & mp.Queue.
+        # Therefore it's important to use torch.Tensor.
+        dataset_dict["image"] = image
+
+        if sem_seg_gt is not None:
+            dataset_dict["sem_seg"] = sem_seg_gt.long()
+
+        if "annotations" in dataset_dict:
+            raise ValueError(
+                "Semantic segmentation dataset should not have 'annotations'."
+            )
+
+        # Prepare per-category binary masks
+        if sem_seg_gt is not None:
+            sem_seg_gt = sem_seg_gt.numpy()
+            instances = Instances(image_shape)
+            classes = np.unique(sem_seg_gt)
+            # remove ignored region
+            classes = classes[classes != self.ignore_label]
+            instances.gt_classes = torch.tensor(classes, dtype=torch.int64)
+
+            masks = []
+            for class_id in classes:
+                masks.append(sem_seg_gt == class_id)
+
+            if len(masks) == 0:
+                # Some image does not have annotation (all ignored)
+                instances.gt_masks = torch.zeros(
+                    (0, sem_seg_gt.shape[-2], sem_seg_gt.shape[-1])
+                )
+            else:
+                masks = BitMasks(
+                    torch.stack(
+                        [
+                            torch.from_numpy(np.ascontiguousarray(x.copy()))
+                            for x in masks
+                        ]
+                    )
+                )
+                instances.gt_masks = masks.tensor
+
+            dataset_dict["instances"] = instances
+
+        return dataset_dict

open_vocab_seg/data/datasets/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Facebook, Inc. and its affiliates.
+from . import register_coco_stuff, register_voc_seg
+from . import register_cc3m
+from . import register_ade20k_full
+from . import register_pascal_context