Update README.md
Browse files
README.md
CHANGED
@@ -78,9 +78,9 @@ pipeline_tag: image-segmentation
|
|
78 |
---
|
79 |
|
80 |
# FLAIR model collection
|
81 |
-
The FLAIR models
|
82 |
The distributed pre-trained models differ in their :
|
83 |
-
* dataset for training : [**FLAIR** dataset] (https://huggingface.co/datasets/IGNF/FLAIR) or the increased version of this dataset **FLAIR-INC** (x 3.5 patches)
|
84 |
* input modalities : **RGB** (natural colours), **RGBI** (natural colours + infrared), **RGBIE** (natural colours + infrared + elevation)
|
85 |
* model architecture : **resnet34_unet** (U-Net with a Resnet-34 encoder), **deeplab**
|
86 |
* target class nomenclature : **12cl** (12 land cover classes) or **15cl** (15 land cover classes)
|
@@ -89,6 +89,7 @@ The distributed pre-trained models differ in their :
|
|
89 |
# FLAIR FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model
|
90 |
|
91 |
The general characteristics of this specific model **FLAIR-INC_RVBIE_resnet34_unet_15cl_norm** are :
|
|
|
92 |
* RGBIE images (true colours + infrared + elevation)
|
93 |
* U-Net with a Resnet-34 encoder
|
94 |
* 15 class nomenclature : [building, pervious_ surface, impervious_surface, bare_soil, water, coniferous, deciduous, brushwood, vineyard, herbaceous, agricultural_land, plowed_land, swimming pool, snow, greenhouse]
|
@@ -122,6 +123,11 @@ _**Multi-domain model**_ :
|
|
122 |
The FLAIR-INC dataset that was used for training is composed of 75 radiometric domains. In the case of aerial images, domain shifts are frequent and are mainly due to : the date of acquisition of the aerial survey (april to november), the spatial domain (equivalent to a french department administrative division) and downstream radimetric processing.
|
123 |
By construction (sampling 75 domains) the model is robust to these shifts, and can be applied to any images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)).
|
124 |
|
|
|
|
|
|
|
|
|
|
|
125 |
_**Land Cover classes of prediction**_ :
|
126 |
The orginial class nomenclature of the FLAIR Dataset is made up of 19 classes(See the [FLAIR dataset](https://huggingface.co/datasets/IGNF/FLAIR) page for details).
|
127 |
However 3 classes corresponding to uncertain labelisation (Mixed (16), Ligneous (17) and Other (19)) and 1 class with very poor labelling (Clear cut (15)) were deasctivated during training.
|
@@ -294,6 +300,8 @@ The following illustration give the confusion matrix :
|
|
294 |
|
295 |
### Results
|
296 |
|
|
|
|
|
297 |
{{ results | default("[More Information Needed]", true)}}
|
298 |
|
299 |
#### Summary
|
@@ -314,10 +322,19 @@ The following illustration give the confusion matrix :
|
|
314 |
|
315 |
**BibTeX:**
|
316 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
317 |
{{ citation_bibtex | default("[More Information Needed]", true)}}
|
318 |
|
319 |
**APA:**
|
320 |
-
|
321 |
{{ citation_apa | default("[More Information Needed]", true)}}
|
322 |
|
323 |
## Contact
|
|
|
78 |
---
|
79 |
|
80 |
# FLAIR model collection
|
81 |
+
The FLAIR models are a collection of semantic segmentation models initially developed to classify land cover on very high resolution aerial images (more specifically the French [BD ORTHO®](https://geoservices.ign.fr/bdortho) product).
|
82 |
The distributed pre-trained models differ in their :
|
83 |
+
* dataset for training : [**FLAIR** dataset] (https://huggingface.co/datasets/IGNF/FLAIR) or the increased version of this dataset **FLAIR-INC** (x 3.5 patches). Only the FLAIR version is open at the moment.
|
84 |
* input modalities : **RGB** (natural colours), **RGBI** (natural colours + infrared), **RGBIE** (natural colours + infrared + elevation)
|
85 |
* model architecture : **resnet34_unet** (U-Net with a Resnet-34 encoder), **deeplab**
|
86 |
* target class nomenclature : **12cl** (12 land cover classes) or **15cl** (15 land cover classes)
|
|
|
89 |
# FLAIR FLAIR-INC_RVBIE_resnet34_unet_15cl_norm model
|
90 |
|
91 |
The general characteristics of this specific model **FLAIR-INC_RVBIE_resnet34_unet_15cl_norm** are :
|
92 |
+
* trained with the FLAIR-INC dataset
|
93 |
* RGBIE images (true colours + infrared + elevation)
|
94 |
* U-Net with a Resnet-34 encoder
|
95 |
* 15 class nomenclature : [building, pervious_ surface, impervious_surface, bare_soil, water, coniferous, deciduous, brushwood, vineyard, herbaceous, agricultural_land, plowed_land, swimming pool, snow, greenhouse]
|
|
|
123 |
The FLAIR-INC dataset that was used for training is composed of 75 radiometric domains. In the case of aerial images, domain shifts are frequent and are mainly due to : the date of acquisition of the aerial survey (april to november), the spatial domain (equivalent to a french department administrative division) and downstream radimetric processing.
|
124 |
By construction (sampling 75 domains) the model is robust to these shifts, and can be applied to any images of the ([BD ORTHO® product](https://geoservices.ign.fr/bdortho)).
|
125 |
|
126 |
+
_**Specification for the Elevation channel**_ :
|
127 |
+
The fifth dimension of the RGBIE images is the Elevation (height of building and vegetation). This information is encoded in a 8-bit encoding format.
|
128 |
+
When decoded to [0,255] ints, a difference of 1 coresponds to 20 cm step of elevation difference.
|
129 |
+
|
130 |
+
|
131 |
_**Land Cover classes of prediction**_ :
|
132 |
The orginial class nomenclature of the FLAIR Dataset is made up of 19 classes(See the [FLAIR dataset](https://huggingface.co/datasets/IGNF/FLAIR) page for details).
|
133 |
However 3 classes corresponding to uncertain labelisation (Mixed (16), Ligneous (17) and Other (19)) and 1 class with very poor labelling (Clear cut (15)) were deasctivated during training.
|
|
|
300 |
|
301 |
### Results
|
302 |
|
303 |
+
<!-- Gio : Add inferenvce Sample ??? -->
|
304 |
+
|
305 |
{{ results | default("[More Information Needed]", true)}}
|
306 |
|
307 |
#### Summary
|
|
|
322 |
|
323 |
**BibTeX:**
|
324 |
|
325 |
+
@misc{garioud2023flair,
|
326 |
+
title={FLAIR: a Country-Scale Land Cover Semantic Segmentation Dataset From Multi-Source Optical Imagery},
|
327 |
+
author={Anatol Garioud and Nicolas Gonthier and Loic Landrieu and Apolline De Wit and Marion Valette and Marc Poupée and Sébastien Giordano and Boris Wattrelos},
|
328 |
+
year={2023},
|
329 |
+
eprint={2310.13336},
|
330 |
+
archivePrefix={arXiv},
|
331 |
+
primaryClass={cs.CV}
|
332 |
+
}
|
333 |
+
|
334 |
{{ citation_bibtex | default("[More Information Needed]", true)}}
|
335 |
|
336 |
**APA:**
|
337 |
+
Garioud, A., Gonthier, N., Landrieu, L., De Wit, A., Valette, M., Poupée, M., ... & Wattrelos, B. (2023). FLAIR: a Country-Scale Land Cover Semantic Segmentation Dataset From Multi-Source Optical Imagery. arXiv preprint arXiv:2310.13336.
|
338 |
{{ citation_apa | default("[More Information Needed]", true)}}
|
339 |
|
340 |
## Contact
|