README.md

---
configs:
- config_name: 'omni'
  data_files:
  - path: data/train/omni/*.tar
    split: train
- config_name: 'cps'
  data_files:
  - path: data/train/cps/*.tar
    split: train
- config_name: 'ycbv'
  data_files:
  - path: data/train/ycbv/*.tar
    split: train
- config_name: 'tless'
  data_files:
  - path: data/train/tless/*.tar
    split: train
task_categories:
- zero-shot-classification
- object-detection
- depth-estimation
- image-classification
- image-segmentation
- image-feature-extraction
- image-to-3d
- zero-shot-object-detection
pretty_name: Dropjects
size_categories:
- 10K<n<1M
---
# Dataset Card for Dropjects

Dropjects is a synthetic stereo RGB-D object dataset, created at the Chair of Cyber-Physical Systems in Production Engineering at the Technical University of Munich.
It contains pose, bounding box, and segmentation masks for different sets of objects.

## Dataset Details

### Subsets
You can choose subsets with different sets of objects. Currently, there are the following subsets/object sets:
- omni (500k images): Contains ~6k objects of the [OmniObject3D dataset](https://omniobject3d.github.io/)
- ycbv (50k images): Contains the [YCB Video objects](https://rse-lab.cs.washington.edu/projects/posecnn/)
- tless (50k images): Contains the [TLESS objects](http://cmp.felk.cvut.cz/t-less/)
- cps (50k images): Contains the Dropjects objects (TBA)

Then you can load the dataset like this, for example all lighting conditions for the stapler in the box, with clutter
```
from datasets import load_dataset

ds = load_dataset("LukasDb/dropjects", "omni", streaming=True, trust_remote_code=True, split='train')
for data in ds.with_format('tensorflow'):
    rgb = data['rgb'] # tf.uint8 Tensor, (h,w,3)
```


### Dataset Description

- **Curated by:** lukas.dirnberger@tum.de
- **License:** CC

### Dataset Sources

<!-- Provide the basic links for the dataset. -->

- **Repository:** TBA
- **Paper:** TBA

## Dataset Structure

TBA

## Citation

**BibTeX:**
TBA

## Dataset Card Authors and Contact

Lukas Dirnberger (lukas.dirnberger@tum.de)