SM737 commited on
Commit
b101522
·
verified ·
1 Parent(s): 3c500e6

Upload folder using huggingface_hub (part 3)

Browse files
Files changed (26) hide show
  1. .gitattributes +3 -0
  2. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/LICENSE +201 -0
  3. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/README.md +575 -0
  4. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/configs/train/config_task.yaml +58 -0
  5. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_FULL.zip +3 -0
  6. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_TOY.zip +3 -0
  7. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/flair-hub-HF-dl.py +849 -0
  8. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/imgs/flair_bandeau.jpg +3 -0
  9. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/setup.py +55 -0
  10. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/__init__.py +0 -0
  11. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/data/dataloader.py +258 -0
  12. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/data/datamodule.py +125 -0
  13. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/main.py +60 -0
  14. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/tasks/stages.py +101 -0
  15. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/tasks/trainers.py +130 -0
  16. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/utils/config_io.py +77 -0
  17. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/utils/messaging.py +114 -0
  18. models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_zonal_detection/main.py +14 -0
  19. models/satquery_water_bundle/MODEL_CARD.md +11 -0
  20. models/satquery_water_bundle/satquery_model_health.json +27 -0
  21. models/satquery_water_bundle/satquery_water_config.json +68 -0
  22. models/satquery_water_bundle/satquery_water_metrics.json +40 -0
  23. models/satquery_water_bundle/satquery_water_state_dict.pt +3 -0
  24. models/satquery_water_bundle/thresholds.json +5 -0
  25. models/water/s2-water-unetplusplus-efficientnet-b4/config.json +9 -0
  26. models/water/s2-water-unetplusplus-efficientnet-b4/model.pth +3 -0
.gitattributes CHANGED
@@ -7,3 +7,6 @@ models/finetuned/landcover/satquery_landcover_v1/confusion_matrix.png filter=lfs
7
  models/finetuned/landcover/satquery_landcover_v1/qualitative_overlays.png filter=lfs diff=lfs merge=lfs -text
8
  models/land_cover/flair-rgb-15cl-deeplabv3/confmat.npy filter=lfs diff=lfs merge=lfs -text
9
  models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/model_utils/AerialROI.png filter=lfs diff=lfs merge=lfs -text
 
 
 
 
7
  models/finetuned/landcover/satquery_landcover_v1/qualitative_overlays.png filter=lfs diff=lfs merge=lfs -text
8
  models/land_cover/flair-rgb-15cl-deeplabv3/confmat.npy filter=lfs diff=lfs merge=lfs -text
9
  models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/model_utils/AerialROI.png filter=lfs diff=lfs merge=lfs -text
10
+ models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_FULL.zip filter=lfs diff=lfs merge=lfs -text
11
+ models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_TOY.zip filter=lfs diff=lfs merge=lfs -text
12
+ models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/imgs/flair_bandeau.jpg filter=lfs diff=lfs merge=lfs -text
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/LICENSE ADDED
@@ -0,0 +1,201 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or a Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding those notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any such Derivative Works as a whole, provided Your use,
127
+ reproduction, and distribution of the Work otherwise complies with
128
+ the conditions stated in this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of
133
+ this License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the content of the NOTICE file.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your exercise of permissions under this License.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as deliberate and grossly
156
+ negligent acts) or agreed to in writing, shall any Contributor be
157
+ liable to You for damages, including any direct, indirect, special,
158
+ incidental, or consequential damages of any character arising as a
159
+ result of this License or out of the use or inability to use the
160
+ Work (including but not limited to damages for loss of goodwill,
161
+ work stoppage, computer failure or malfunction, or any and all
162
+ other commercial damages or losses), even if such Contributor
163
+ has been advised of the possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on Your own behalf and on Your sole responsibility, not on behalf
171
+ of any other Contributor, and only if You agree to indemnify,
172
+ defend, and hold each Contributor harmless for any liability
173
+ incurred by, or claims asserted against, such Contributor by reason
174
+ of your accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ APPENDIX: How to apply the Apache License to your work.
179
+
180
+ To apply the Apache License to your work, attach the following
181
+ boilerplate notice, with the fields enclosed by brackets "[]"
182
+ replaced with your own identifying information. (Don't include
183
+ the brackets!) The text should be enclosed in the appropriate
184
+ comment syntax for the file format. We also recommend that a
185
+ file or class name and description of purpose be included on the
186
+ same "printed page" as the copyright notice for easier
187
+ identification within third-party archives.
188
+
189
+ Copyright [yyyy] [name of copyright owner]
190
+
191
+ Licensed under the Apache License, Version 2.0 (the "License");
192
+ you may not use this file except in compliance with the License.
193
+ You may obtain a copy of the License at
194
+
195
+ http://www.apache.org/licenses/LICENSE-2.0
196
+
197
+ Unless required by applicable law or agreed to in writing, software
198
+ distributed under the License is distributed on an "AS IS" BASIS,
199
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200
+ See the License for the specific language governing permissions and
201
+ limitations under the License.
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/README.md ADDED
@@ -0,0 +1,575 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <div align="center">
2
+
3
+ # FLAIR-HUB
4
+ # Multimodal & multitask semantic segmentation of Earth Observation imagery
5
+
6
+
7
+ ![Static Badge](https://img.shields.io/badge/Code%3A-lightgrey?color=lightgrey) [![license](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://github.com/IGNF/FLAIR-HUB/blob/master/LICENSE) <a href="https://pytorch.org/get-started/locally/"><img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-ee4c2c?logo=pytorch&logoColor=white"></a>
8
+ <a href="https://pytorchlightning.ai/"><img alt="Lightning" src="https://img.shields.io/badge/-Lightning-792ee5?logo=pytorchlightning&logoColor=white"></a> &emsp; ![Static Badge](https://img.shields.io/badge/Dataset%3A-lightgrey?color=lightgrey) [![license](https://img.shields.io/badge/License-IO%202.0-green.svg)](https://github.com/etalab/licence-ouverte/blob/master/open-licence.md)
9
+
10
+
11
+
12
+
13
+ Participate in obtaining more accurate maps for a more comprehensive description and a better understanding of our environment! Come push the limits of state-of-the-art semantic segmentation approaches on a large and challenging dataset. <br> Get in touch at :email: flair@ign.fr
14
+
15
+
16
+
17
+ ![Alt bandeau FLAIR-HUB](imgs/flair_bandeau.jpg?raw=true)
18
+
19
+ </div>
20
+
21
+ <div style="border-width:1px; border-style:solid; border-color:#d2db8c; padding-left: 1em; padding-right: 1em; ">
22
+
23
+ <h2 style="margin-top:5px;">:mag_right:Quicklinks</h2>
24
+
25
+
26
+ - **Entry page :** [Overall FLAIR-HUB page](https://ignf.github.io/FLAIR/FLAIR-HUB/flairhub.html) <br>
27
+
28
+ - **Datapaper :** [ISPRS Journal](https://www.sciencedirect.com/science/article/pii/S0924271626001899) / [arXiv Preprint](https://arxiv.org/abs/2506.07080)
29
+
30
+ - **Dataset links :** [Dataset link](https://huggingface.co/datasets/IGNF/FLAIR-HUB)
31
+
32
+ - **Pre-trained models :** [Pretrained models](https://huggingface.co/collections/IGNF/flair-models-684035e78bd5bff99199ff87)
33
+
34
+ - **Toy Datatset direct downlaod** [Dataset link](https://storage.gra.cloud.ovh.net/v1/AUTH_366279ce616242ebb14161b7991a8461/defi-ia/flair_hub/FLAIR-HUB_TOY_DATASET.zip)
35
+
36
+ </div>
37
+ <hr>
38
+
39
+ ## Context and data
40
+
41
+ This repository is built on top of the **FLAIR-HUB** dataset, the largest multi-sensor land cover dataset with very-high-resolution (20\:cm) annotations, covering 2528\:km² of France. It combines six aligned modalities: aerial imagery, Sentinel-1/2 time series, SPOT imagery, topographic data, and historical aerial images. Extensive benchmarks evaluate multimodal fusion and deep learning models (CNNs, transformers) for land cover or crop mapping and also explore multi-task learning. Key figures of the dataset are : <br>
42
+
43
+ <center>
44
+ <table style="width:95%;max-width:600px;">
45
+
46
+ <tbody>
47
+ <tr><td>🗺️</td><td>ROI / Area Covered</td><td style="text-align: center">2,822 ROIs / 2,528 km²</td></tr>
48
+ <tr><td>🏛️</td><td>Departments (France)</td><td style="text-align: center">74</td></tr>
49
+ <tr><td>🧩</td><td>AI Patches (512×512 px)</td><td style="text-align: center">241,100</td></tr>
50
+ <tr><td>🖼️</td><td>Annotated Pixels</td><td style="text-align: center">63.2 billion</td></tr>
51
+ <tr><td>🛰️</td><td>Sentinel-2 Acquisitions</td><td style="text-align: center">256,221</td></tr>
52
+ <tr><td>📡</td><td>Sentinel-1 Acquisitions</td><td style="text-align: center">532,696</td></tr>
53
+ <tr><td>📁</td><td>Total Files</td><td style="text-align: center">~2.5 million</td></tr>
54
+ <tr><td>💾</td><td>Total Dataset Size</td><td style="text-align: center">~750 GB</td></tr>
55
+ </tbody>
56
+ </table>
57
+ </center>
58
+
59
+ <hr>
60
+ <br><br>
61
+
62
+ ## Installation
63
+
64
+ 1. Conda env
65
+ ```
66
+ $ conda create -n FLAIRHUB python=3.11 pip -y
67
+ ```
68
+
69
+ 2. Force torch with GPU support
70
+ ```
71
+ $ conda activate FLAIRHUB
72
+ $ pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --extra-index-url https://download.pytorch.org/whl/cu126
73
+ ```
74
+ 3. Install flair-hub
75
+ ```
76
+ $ pip install -e .
77
+ ```
78
+
79
+ <hr>
80
+ <br><br>
81
+
82
+ ## Entry points :mag_right:
83
+
84
+ flair-hub has two entry points:
85
+
86
+ #### :file_folder: src/flair_hub --> used to train, infer models, calculate metrics, at the patch level. <br>
87
+ ```
88
+ $ flairhub --config /path/configs/train/
89
+ ```
90
+
91
+ #### :file_folder: src/flair_zonal_detection --> used to infer a pretrained model over larger areas. <br>
92
+ ```
93
+ $ flairhub_zonal --config /path/configs/config_zonal_detection.yaml
94
+ ```
95
+
96
+
97
+ <hr>
98
+ <br><br>
99
+
100
+ ## Configuration for flair-hub :page_facing_up:
101
+
102
+ The pipeline takes as input a folder with 4 configuration YAML files or a single YAML file contaning all arguments. The configuration file includes sections for data paths, tasks, supervision, model configuration, hyperparameters and computational resources.
103
+
104
+ #### config_task.yaml : <br>
105
+ ```yaml
106
+ paths:
107
+ out_folder: Directory to store all output results (models, logs, predictions).
108
+ out_model_name: Name identifier for the saved model.
109
+ train_csv: CSV files containing paths and labels for training datasets.
110
+ val_csv: CSV files containing paths and labels for validation datasets.
111
+ test_csv: CSV files containing paths and labels for test datasets.
112
+ global_mtd_folder: Path to folder with global metadata required for processing.
113
+ ckpt_model_path: Path to a pretrained model checkpoint to initialize weights.
114
+
115
+ tasks:
116
+ train: Enable or disable training phase.
117
+ train_tasks:
118
+ init_weights_only_from_ckpt: Use checkpoint weights for initialization only (no training resume).
119
+ resume_training_from_ckpt: Resume training from the provided checkpoint.
120
+ predict: Run model inference on test set.
121
+ write_files: Save predictions and outputs to disk.
122
+ georeferencing_output: Apply georeferencing to output files.
123
+ metrics_only: Compute and report metrics without running training or prediction.
124
+
125
+ hyperparams:
126
+ num_epochs: Total number of training epochs.
127
+ batch_size: Number of samples per training batch.
128
+ seed: Random seed for reproducibility.
129
+ learning_rate: Initial learning rate for optimizer.
130
+ optimizer: Choice of optimizer (adamw, adam, or sgd).
131
+ optim_weight_decay: Weight decay regularization.
132
+ optim_betas: Beta coefficients for Adam-based optimizers.
133
+ scheduler: Learning rate scheduler strategy.
134
+ warmup_fraction: Warm-up fraction for one_cycle_lr.
135
+ plateau_patience: Patience for reduce_on_plateau scheduler.
136
+
137
+ hardware:
138
+ accelerator: Type of hardware to use (gpu, cpu).
139
+ num_nodes: Number of nodes for distributed training.
140
+ gpus_per_node: GPUs used per node.
141
+ strategy: Distributed training strategy (e.g., auto, ddp).
142
+ num_workers: Number of data loading workers.
143
+
144
+ saving:
145
+ ckpt_save_also_last: Save the final model checkpoint at the end of training.
146
+ ckpt_weights_only: Save only the model weights (not optimizer or scheduler).
147
+ ckpt_monitor: Metric used to determine the best checkpoint.
148
+ ckpt_monitor_mode: Direction to optimize the monitored metric (max or min).
149
+ ckpt_earlystopping_patience: Epochs to wait for improvement before stopping.
150
+ cp_csv_and_conf_to_output: Copy configuration and CSV files to output directory.
151
+ enable_progress_bar: Show training progress bar.
152
+ progress_rate: Update frequency of progress bar (in steps).
153
+ ckpt_verbose: Print detailed checkpoint saving information.
154
+ verbose_config: Print full config details to console.
155
+ ```
156
+
157
+ #### config_modalities.yaml : <br>
158
+ ```yaml
159
+ modalities:
160
+
161
+ inputs:
162
+ AERIAL_RGBI : Enable/disable AERIAL_RGBI modality. [True/False]
163
+ AERIAL-RLT_PAN : [True/False]
164
+ DEM_ELEV : [True/False]
165
+ SPOT_RGBI : [True/False]
166
+ SENTINEL2_TS : [True/False].
167
+ SENTINEL1-ASC_TS : [True/False]
168
+ SENTINEL1-DESC_TS : [True/False]
169
+
170
+ inputs_channels:
171
+ AERIAL_RGBI : Selected channels for AERIAL_RGBI input. Starts at 1.
172
+ SPOT_RGBI : Selected channels for SPOT_RGBI input. Starts at 1.
173
+ SENTINEL2_TS : Selected channels for Sentinel-2 time series. Starts at 1.
174
+ SENTINEL1-ASC_TS : Selected channels for Sentinel-1 ascending time series. Starts at 1.
175
+ SENTINEL1-DESC_TS : Selected channels for Sentinel-1 descending time series. Starts at 1.
176
+
177
+ aux_loss:
178
+ AERIAL_RGBI : Apply auxiliary loss to AERIAL_RGBI input if mutliple modalities. [True/False]
179
+ AERIAL-RLT_PAN : [True/False]
180
+ DEM_ELEV : [True/False]
181
+ SPOT_RGBI : [True/False]
182
+ SENTINEL2_TS : [True/False]
183
+ SENTINEL1-ASC_TS : [True/False]
184
+ SENTINEL1-DESC_TS : [True/False]
185
+
186
+ aux_loss_weight: Multiplier for auxiliary loss before combining with main loss. Default 1.
187
+ modality_dropout: Dropout probability per modality (0 = keep all, 1 = drop all systematically).
188
+ AERIAL_RGBI : 0
189
+ AERIAL-RLT_PAN : 0
190
+
191
+ pre_processings:
192
+ filter_sentinel2: Enable filtering of Sentinel-2 based on masks.
193
+ filter_sentinel2_max_cloud : Max acceptable cloud cover in Sentinel-2 [%]. Default 1.
194
+ filter_sentinel2_max_snow : Max acceptable snow cover in Sentinel-2 [%]. Default 1.
195
+ filter_sentinel2_max_frac_cover : Max acceptable fractional invalid coverage in Sentinel-2. Default 0.05.
196
+ temporal_average_sentinel2 : Temporal averaging method for Sentinel-2 (False/monthly/semi-monthly).
197
+ temporal_average_sentinel1 : Temporal averaging method for Sentinel-1 (False/monthly/semi-monthly).
198
+ calc_elevation : Enable calculation oF DTM-DSM.
199
+ calc_elevation_stack_dsm : Stack DSM if elevation calculated.
200
+ use_augmentation: Enable data augmentation for training. Default False.
201
+
202
+ normalization:
203
+ norm_type : Normalization strategy (custom, scaling, without).
204
+ AERIAL_RGBI_means : Mean values used to normalize AERIAL_RGBI input if custom.
205
+ AERIAL_RGBI_stds : Std. dev. values used to normalize AERIAL_RGBI input if custom.
206
+ AERIAL-RLT_PAN_means : Mean values for AERIAL-RLT_PAN input if custom.
207
+ AERIAL-RLT_PAN_stds : Std. dev. values for AERIAL-RLT_PAN input if custom.
208
+ SPOT_RGBI_means : Mean values for SPOT_RGBI input if custom.
209
+ SPOT_RGBI_stds : Std. dev. values for SPOT_RGBI input if custom.
210
+ DEM_ELEV_means : Mean elevation values (DSM, DTM) if custom.
211
+ DEM_ELEV_stds : Std. dev. of elevation values (DSM, DTM) if custom.
212
+ ```
213
+
214
+ #### config_models.yaml : <br>
215
+ ```yaml
216
+ models:
217
+
218
+ monotemp_model: # Encoder-decoder architecture for single-date (monotemporal) inputs
219
+
220
+ arch: Encoder-decoder architecture from SMP. Eg, swin_tiny_patch4_window7_224-upernet
221
+ new_channels_init_mode: Strategy to initialize new input channels if above 3 channels (options: copy_first, copy_second, copy_third, random).
222
+
223
+ multitemp_model: # Architecture for handling multi-temporal (time series) inputs
224
+
225
+ ref_date: Reference date (MM-DD) for temporal position encoding.
226
+ encoder_widths: Channel sizes at each stage in the encoder; must match decoder.
227
+ decoder_widths: Channel sizes at each stage in the decoder; must match encoder.
228
+ out_conv: Output convolutional layer configuration [intermediate_channels, num_classes].
229
+ str_conv_k: Kernel size for structured convolutions.
230
+ str_conv_s: Stride for structured convolutions.
231
+ str_conv_p: Padding for structured convolutions.
232
+ agg_mode: Aggregation method for temporal features (e.g., att_group).
233
+ encoder_norm: Normalization type used in the encoder (e.g., group).
234
+ n_head: Number of attention heads in the model.
235
+ d_model: Dimensionality of the model features.
236
+ d_k: Key dimensionality for attention mechanism.
237
+ pad_value: Value used for padding missing temporal data.
238
+ padding_mode: Type of padding used in convolutions (e.g., reflect).
239
+ ```
240
+
241
+ #### config_supervision.yaml : <br>
242
+ ```yaml
243
+ labels: # List of tasks
244
+ - AERIAL_LABEL-COSIA
245
+ - ALL_LABEL-LPIS
246
+
247
+ labels_configs:
248
+
249
+ AERIAL_LABEL-COSIA:
250
+ task_weight: Weight assigned to this task during multi-task training. Default 1.
251
+ value_name: Mapping of class indices to semantic category names.
252
+ 0 : 'building'
253
+ 1 : 'greenhouse'
254
+ 2 : 'swimming_pool'
255
+ 3 : 'impervious surface'
256
+ 4 : 'pervious surface'
257
+ 5 : 'bare soil'
258
+ 6 : 'water'
259
+ 7 : 'snow'
260
+ 8 : 'herbaceous vegetation'
261
+ 9 : 'agricultural land'
262
+ 10 : 'plowed land'
263
+ 11 : 'vineyard'
264
+ 12 : 'deciduous'
265
+ 13 : 'coniferous'
266
+ 14 : 'brushwood'
267
+ 15 : 'clear cut'
268
+ 16 : 'ligneous'
269
+ 17 : 'mixed'
270
+ 18 : 'undefined'
271
+ value_weights:
272
+ default: Default weight for all classes (used in loss calculation). Default 1.
273
+ default_exceptions: Specific class indices assigned 0 weight (ignored).
274
+ 15: 0
275
+ 16: 0
276
+ 17: 0
277
+ 18: 0
278
+ per_modality_exceptions: Placeholder for setting modality-specific class weights.
279
+ AERIAL_RGBI:
280
+ 18: 0
281
+
282
+ ALL_LABEL-LPIS:
283
+ task_weight: Weight assigned to this task during multi-task training.
284
+ label_channel_nomenclature: Index of the label channel used.
285
+ value_name: Mapping of class indices to crop type names.
286
+ value_weights:
287
+ default: Default class weight for the entire label set.
288
+ default_exceptions: No exception weights defined (can be added).
289
+ per_modality_exceptions: Placeholder for modality-specific class weight overrides.
290
+ ```
291
+ <br><br>
292
+
293
+
294
+ ### Configuration for flair_zonal_detection :page_facing_up:
295
+
296
+ #### config_zonal_detection.yaml : <br>
297
+ ```yaml
298
+ # ======================
299
+ # I/O
300
+ # ======================
301
+ output_path: Path to store output results.
302
+ output_name: Identifier name for the output files.
303
+
304
+ write_dataframe: Whether to write slicing geometries of input area. [True/False]
305
+ output_type: Format of the output predictions (argmax or class_prob).
306
+ cog_conversion: Convert outputs to Cloud-Optimized GeoTIFF format. [True/False]
307
+
308
+ # ======================
309
+ # Model & Inference
310
+ # ======================
311
+ model_weights: Path to pretrained model weights for inference. [.safetensors / .ckpt / .pth]
312
+ use_gpu: Whether to use GPU for inference.
313
+ batch_size: Number of samples per inference batch.
314
+ num_worker: Number of data loading workers during inference.
315
+
316
+ img_pixels_detection: Size of input image chunks in pixels.
317
+ margin: Overlap margin (in pixels) between tiles.
318
+ output_px_meters: Output resolution in meters per pixel.
319
+
320
+ # ======================
321
+ # Model Framework
322
+ # ======================
323
+ monotemp_arch: SMP Architecture used for monotemporal inference. Eg., swin_base_patch4_window12_384-upernet
324
+ multitemp_model_ref_date: Reference date (MM-DD) for aligning multi-temporal inputs.
325
+
326
+ # ======================
327
+ # Modalities (inputs used for inference)
328
+ # ======================
329
+ modalities:
330
+
331
+ inputs:
332
+ AERIAL_RGBI : Enable AERIAL_RGBI input. [True / False]
333
+ AERIAL-RLT_PAN : [True / False]
334
+ DEM_ELEV : [True / False]
335
+ SPOT_RGBI : [True / False]
336
+ SENTINEL2_TS : [True / False]
337
+ SENTINEL1-ASC_TS : [True / False]
338
+ SENTINEL1-DESC_TS : [True / False]
339
+
340
+ AERIAL_RGBI:
341
+ input_img_path: Path to AERIAL_RGBI input image.
342
+ channels: Channels used from the input image. Starts at 1.
343
+ normalization:
344
+ type: Normalization method (e.g., custom).
345
+ means: Per-channel mean for normalization.
346
+ stds: Per-channel standard deviation for normalization.
347
+
348
+ AERIAL-RLT_PAN:
349
+ input_img_path: Path to AERIAL-RLT_PAN input.
350
+ channels: Channels used.
351
+ normalization:
352
+ type: Normalization method.
353
+ means: Mean value.
354
+ stds: Standard deviation.
355
+
356
+ DEM_ELEV:
357
+ input_img_path: Path to DEM input.
358
+ channels: Channels used.
359
+ normalization:
360
+ type: Normalization method.
361
+ means: Per-channel mean.
362
+ stds: Per-channel standard deviation.
363
+ calc_elevation: Whether to compute elevation from DSM/DTM.
364
+ calc_elevation_stack_dsm: Stack derived DSM with the input.
365
+
366
+ SPOT_RGBI:
367
+ input_img_path: Path to SPOT imagery input.
368
+ channels: Channels used.
369
+ normalization:
370
+ type: Normalization method.
371
+ means: Per-channel mean.
372
+ stds: Per-channel standard deviation.
373
+
374
+ SENTINEL2_TS:
375
+ input_img_path: Path to Sentinel-2 time series input.
376
+ channels: Channels used.
377
+ dates_txt: Path to text file listing image dates.
378
+ filter_clouds: Enable cloud filtering.
379
+ filter_clouds_img_path: Path to Sentinel-2 cloud mask.
380
+ temporal_average: Whether to average the time series over time.
381
+
382
+ SENTINEL1-ASC_TS:
383
+ input_img_path: Path to Sentinel-1 ascending orbit input.
384
+ channels: Channels used.
385
+ dates_txt: Path to date list (optional).
386
+ temporal_average: Whether to average over time.
387
+
388
+ SENTINEL1-DESC_TS:
389
+ input_img_path: Path to Sentinel-1 descending orbit input.
390
+ channels: Channels used.
391
+ dates_txt: Path to date list (optional).
392
+ temporal_average: Whether to average over time.
393
+
394
+ # ======================
395
+ # Tasks
396
+ # ======================
397
+ tasks:
398
+
399
+ - name: AERIAL_LABEL-COSIA # Semantic segmentation label task (land cover classes)
400
+ active: Whether to activate this label set during inference. [True / False]
401
+ class_names: Mapping of class indices to category names.
402
+ 0: building
403
+ 1: greenhouse
404
+ 2: swimming_pool
405
+ 3: impervious surface
406
+ 4: pervious surface
407
+ 5: bare soil
408
+ 6: water
409
+ 7: snow
410
+ 8: herbaceous vegetation
411
+ 9: agricultural land
412
+ 10: plowed land
413
+ 11: vineyard
414
+ 12: deciduous
415
+ 13: coniferous
416
+ 14: brushwood
417
+ 15: clear cut
418
+ 16: ligneous
419
+ 17: mixed
420
+ 18: undefined
421
+
422
+ - name: ALL_LABEL-LPIS # Crop type classification label task
423
+ active: Whether to activate this label set during inference. [True / False]
424
+ class_names: Mapping of class indices to crop categories.
425
+ ```
426
+
427
+
428
+
429
+ <hr>
430
+
431
+ <br><br>
432
+
433
+ ## Baseline results
434
+
435
+ | Model ID | Aerial VHR | Elevation | SPOT | S2 t.s. | S1 t.s. | Historical | PARA. | O.A. | mIoU |
436
+ |----------|------------|-----------|------|---------|---------|------------|--------|------|------|
437
+ |||||||||||
438
+ | LC-A | ✓ | | | | | | 89.4 | 77.5 | 64.1 |
439
+ | LC-B | ✓ | ✓ | | | | | 181.4 | 78.1 | 65.1 |
440
+ | LC-C | ✓ | ✓ | ✓ | | | | 270.6 | 78.2 | 65.2 |
441
+ | LC-D | ✓ | | | ✓ | | | 93.9 | 77.6 | 64.7 |
442
+ | LC-E | ✓ | | | | ✓ | | 95.8 | 77.7 | 64.5 |
443
+ | LC-F | ✓ | | | ✓ | ✓ | | 97.7 | 77.7 | 64.9 |
444
+ | LC-G | | | | ✓ | | | 0.9 | 57.8 | 34.2 |
445
+ | LC-H | | | | | ✓ | | 1.8 | 54.5 | 28.2 |
446
+ | LC-I | | | ✓ | | | | 89.2 | 64.1 | 43.5 |
447
+ | LC-J | | ✓ | | | | | 89.4 | 67.4 | 51.2 |
448
+ | LC-K | ✓ | | | | | ✓ | 181.4 | 77.6 | 64.3 |
449
+ | LC-L | ✓ | ✓ | ✓ | ✓ | ✓ | | 276.4 | 78.2 | 65.8 |
450
+ | LC-ALL | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 365.8 | 78.2 | 65.6 |
451
+ |||||||||||
452
+ | LPIS-A | ✓ | | | | | | 89.4 | 86.6 | 24.4 |
453
+ | LPIS-B | ✓ | | ✓ | | | | 181.2 | 87.1 | 26.1 |
454
+ | LPIS-C | ✓ | | | ✓ | | | 93.9 | 87.5 | 29.8 |
455
+ | LPIS-D | ✓ | | | ✓ | ✓ | | 97.7 | 88.0 | 36.1 |
456
+ | LPIS-E | ✓ | | ✓ | ✓ | | | 183.1 | 87.6 | 30.3 |
457
+ | LPIS-F | | | | ✓ | | | 0.9 | 85.3 | 23.8 |
458
+ | LPIS-G | | | | | ✓ | | 1.8 | 84.5 | 18.1 |
459
+ | LPIS-H | | | | ✓ | ✓ | | 2.8 | 84.9 | 23.8 |
460
+ | LPIS-I | | | ✓ | ✓ | ✓ | | 97.5 | 87.2 | 39.2 |
461
+ | LPIS-J | ✓ | | ✓ | ✓ | ✓ | | 186.9 | 88.0 | 35.4 |
462
+ | LPIS-K | | | ✓ | | | | 89.2 | 84.5 | 15.1 |
463
+
464
+
465
+
466
+
467
+ ## Inference classes and colors
468
+
469
+ ```yaml
470
+ flair_LC =
471
+ 0 : 'building' 'bâtiment' '#db0e9a'
472
+ 1 : 'greenhouse' 'serre' '#9999ff'
473
+ 2 : 'swimming_pool' 'piscine' '#3de6eb'
474
+ 3 : 'impervious surface' 'zone imperméable' '#f80c00'
475
+ 4 : 'pervious surface' 'zone perméable' '#938e7b'
476
+ 5 : 'bare soil' 'sol nu' '#a97101'
477
+ 6 : 'water' 'eau' '#1553ae'
478
+ 7 : 'snow' 'neige' '#ffffff'
479
+ 8 : 'herbaceous vegetation' 'surface herbacée' '#55ff00'
480
+ 9 : 'agricultural land' 'culture' '#fff30d'
481
+ 10 : 'plowed land' 'terre labourée' '#e4df7c'
482
+ 11 : 'vineyard' 'vigne' '#660082'
483
+ 12 : 'deciduous' 'feuillu' '#46e483'
484
+ 13 : 'coniferous' 'conifère' '#194a26'
485
+ 14 : 'brushwood' 'broussaille' '#f3a60d'
486
+ 15 : 'clear cut' 'coupe' '#8ab3a0'
487
+ 16 : 'ligneous' 'ligneux' '#c5dc42'
488
+ 17 : 'mixed' 'mixte' '#6b714f'
489
+ 18 : 'undefined' 'indéterminé' '#000000'
490
+
491
+ flair_LPIS =
492
+ 0 : 'Grasses' 'Herbe prédominante' '#92d050'
493
+ 1 : 'Wheat' 'Blé' '#d7e600'
494
+ 2 : 'Barley' 'Orge' '#e0e000'
495
+ 3 : 'Maize' 'Maïs' '#fff100'
496
+ 4 : 'Other cereals' 'Autres céréales' '#ffff00'
497
+ 5 : 'Rice' 'Riz' '#e8e8e8'
498
+ 6 : 'Hemp/Flax/Tobacco' 'Lin/Chanvre/Tabac' '#dceaf7'
499
+ 7 : 'Sunflower' 'Tournesol' '#d29ead'
500
+ 8 : 'Rapeseed' 'Colza' '#d29ed0'
501
+ 9 : 'Other oilseed crops' 'Autres oléagineux' '#ffbe99'
502
+ 10 : 'Soy' 'Soja' '#ffc000'
503
+ 11 : 'Other protein crops' 'Autres protéagineux' '#ff9000'
504
+ 12 : 'Fodder legumes' 'Légumineuse fourragères' '#009999'
505
+ 13 : 'Beetroots' 'Betterave' '#808000'
506
+ 14 : 'Potatoes' 'Pomme de terre' '#a7a700'
507
+ 15 : 'Other arable crops' 'Autres arables' '#89896d'
508
+ 16 : 'Vineyard' 'Vigne' '#f2cfee'
509
+ 17 : 'Olive groves' 'Oliveraie' '#6f6633'
510
+ 18 : 'Fruit orchards' 'Verger fruits' '#ac8141'
511
+ 19 : 'Nut orchards' 'Verger noix' '#996633'
512
+ 20 : 'Other permanent crops' 'Autres pérenne' '#80c1d7'
513
+ 21 : 'Mixed crops' 'Mélange' '#000000'
514
+ 22 : 'Background' 'Autres' '#000000'
515
+
516
+ ```
517
+
518
+
519
+
520
+
521
+ ## 🔎 Filter dataset with the FLAIR-HUB Dataset Browser
522
+
523
+ A small desktop GUI to browse and download subsets
524
+ of the IGNF/FLAIR-HUB dataset from Hugging Face with filters for: Domain, Year, Modality or Data type.
525
+
526
+ Requirements:
527
+
528
+ - Python 3.9+
529
+ - Tkinter (usually included; on Linux you may need: sudo apt-get install python3-tk)
530
+ - Python packages: pip install `huggingface_hub`
531
+
532
+ Run:
533
+ 1. Get the `flair-hub-HF-dl.py` file.
534
+ 2. In a terminal: ```pip install huggingface_hub```
535
+ 3. Launch: ```python flair-hub-HF-dl.py```
536
+
537
+ <br><br>
538
+
539
+
540
+ ## Reference
541
+ Cite the following article if you use the FLAIR #1 dataset:
542
+
543
+ ```bibtex
544
+ @article{GARIOUD2026271,
545
+ title = {FLAIR-HUB: Large-scale multimodal dataset for land cover and crop mapping},
546
+ author = {Anatol Garioud and Sébastien Giordano and Nicolas David and Nicolas Gonthier},
547
+ journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
548
+ volume = {237},
549
+ pages = {271-300},
550
+ year = {2026},
551
+ issn = {0924-2716},
552
+ doi = {https://doi.org/10.1016/j.isprsjprs.2026.04.017},
553
+ url = {https://www.sciencedirect.com/science/article/pii/S0924271626001899},
554
+ }
555
+ ```
556
+
557
+ ```
558
+ Anatol Garioud, Sébastien Giordano, Nicolas David, Nicolas Gonthier.
559
+ FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mapping.
560
+ ISPRS Journal of Photogrammetry and Remote Sensing, Volume 237, 2026.
561
+ DOI: https://doi.org/10.1016/j.isprsjprs.2026.04.017
562
+ ```
563
+
564
+
565
+
566
+
567
+ ## Acknowledgment
568
+ The experiments conducted in this study were performed using HPC/AI resources provided by GENCI-IDRIS (Grant 2024-A0161013803, 2024-AD011014286R2 and 2025-A0181013803).
569
+
570
+ ## Dataset license
571
+
572
+ The "OPEN LICENCE 2.0/LICENCE OUVERTE" is a license created by the French government specifically for the purpose of facilitating the dissemination of open data by public administration.
573
+ If you are looking for an English version of this license, you can find it on the official GitHub page at the [official github page](https://github.com/etalab/licence-ouverte).<br>
574
+ Applicable legislation : This licence is governed by French law.<br>
575
+ Compatibility of this licence : This licence has been designed to be compatible with any free licence that at least requires an acknowledgement of authorship, and specifically with the previous version of this licence as well as with the following licences: United Kingdom’s “Open Government Licence” (OGL), Creative Commons’ “Creative Commons Attribution” (CC-BY) and Open Knowledge Foundation’s “Open Data Commons Attribution” (ODC-BY).
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/configs/train/config_task.yaml ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ paths :
2
+
3
+ out_folder: ../results/
4
+ out_model_name: test-lib
5
+
6
+ train_csv: ../toy/TRAIN_FLAIR-INC_TOY.csv
7
+ val_csv: ../toy/VALID_FLAIR-INC_TOY.csv
8
+ test_csv: ../toy/TEST_FLAIR-INC_TOY.csv
9
+ global_mtd_folder: FLAIR-HUB/GLOBAL_ALL_MTD/
10
+ ckpt_model_path:
11
+
12
+
13
+ tasks:
14
+
15
+ train: True
16
+ train_tasks:
17
+ init_weights_only_from_ckpt: False
18
+ resume_training_from_ckpt: False
19
+
20
+ predict: True
21
+ write_files: True
22
+ georeferencing_output: True
23
+
24
+ metrics_only: False
25
+
26
+ hyperparams:
27
+
28
+ num_epochs: 150
29
+ batch_size: 5
30
+ seed: 2025
31
+ learning_rate: 0.00005
32
+
33
+ optimizer: adamw # [sgd / adam / adamw]
34
+ optim_weight_decay: 0.01
35
+ optim_betas: [0.9, 0.999]
36
+
37
+ scheduler: one_cycle_lr # [one_cycle_lr / reduce_on_plateau / cycle_then_plateau]
38
+ warmup_fraction: 0.2 # if using one_cycle_lr. [0-1]
39
+ plateau_patience: 10 # if using cycle_then_plateau
40
+
41
+ hardware:
42
+ accelerator: 'gpu'
43
+ num_nodes: 1
44
+ gpus_per_node: 1
45
+ strategy: 'auto' # ddp_find_unused_parameters_true if num_nodes > 1
46
+ num_workers: 0
47
+
48
+ saving:
49
+ ckpt_save_also_last: True
50
+ ckpt_weights_only: False
51
+ ckpt_monitor: 'val_miou'
52
+ ckpt_monitor_mode: 'max'
53
+ ckpt_earlystopping_patience: 20
54
+ cp_csv_and_conf_to_output: True
55
+ enable_progress_bar: True
56
+ progress_rate: 10
57
+ ckpt_verbose: True
58
+ verbose_config: False
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_FULL.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b903a51a83157e4637d951f80101d2e3a70cdfef0fe4cb819948777ba5c1c6d7
3
+ size 8803545
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/csv/FLAIR-HUB_TOY.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f719cf0a6db767679d2c16323d7d78fb571d2d9cad23298ddfa1bb6b4e171715
3
+ size 18549
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/flair-hub-HF-dl.py ADDED
@@ -0,0 +1,849 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import time
3
+ import threading
4
+ import tkinter as tk
5
+ from tkinter import ttk, messagebox, filedialog
6
+ from huggingface_hub import HfApi, hf_hub_download
7
+
8
+ # ============================================================
9
+ # CONFIG
10
+ # ============================================================
11
+
12
+ DATASET_ID = "IGNF/FLAIR-HUB"
13
+ DEFAULT_DOWNLOAD_DIR = "./FLAIR-HUB_download"
14
+
15
+ os.environ["HF_HUB_DISABLE_PROGRESS_BARS"] = "1"
16
+
17
+ api = HfApi()
18
+
19
+ checkbox_state = {}
20
+ file_sizes = {}
21
+ stop_flag = threading.Event()
22
+ dataset_files = []
23
+
24
+
25
+ # ============================================================
26
+ # HELPER FUNCTIONS
27
+ # ============================================================
28
+
29
+ def human_bytes(num):
30
+ if num is None:
31
+ return "unknown"
32
+ for unit in ["B", "KB", "MB", "GB", "TB"]:
33
+ if num < 1024:
34
+ return f"{num:.1f} {unit}"
35
+ num /= 1024
36
+ return f"{num:.1f} PB"
37
+
38
+
39
+ def parse_zip_metadata(path: str):
40
+ """
41
+ Parse data ZIP filename of the form:
42
+ DOMAIN-YEAR_MODALITY_DATATYPE.zip
43
+
44
+ Example:
45
+ data/D004-2021_AERIAL_RGBI_IMS.zip
46
+ -> domain=D004, year=2021, modality=AERIAL_RGBI, datatype=IMS
47
+ """
48
+ name = os.path.basename(path)
49
+ if not name.lower().endswith(".zip"):
50
+ return None
51
+
52
+ if name == "GLOBAL_ALL_MTD.zip":
53
+ return None
54
+
55
+ stem = name[:-4]
56
+ if "-" not in stem:
57
+ return None
58
+
59
+ domain, rest = stem.split("-", 1)
60
+ parts = rest.split("_")
61
+ year = parts[0] if parts and parts[0].isdigit() else None
62
+
63
+ modality = None
64
+ datatype = None
65
+
66
+ if len(parts) >= 2:
67
+ if len(parts) >= 3:
68
+ datatype = parts[-1]
69
+ modality = "_".join(parts[1:-1])
70
+ else:
71
+ modality = parts[1]
72
+
73
+ return {
74
+ "path": path,
75
+ "name": name,
76
+ "domain": domain,
77
+ "year": year,
78
+ "modality": modality,
79
+ "datatype": datatype,
80
+ }
81
+
82
+
83
+ # ============================================================
84
+ # MAIN WINDOW & STYLES
85
+ # ============================================================
86
+
87
+ root = tk.Tk()
88
+ root.title("FLAIR-HUB Dataset Download Tool")
89
+ root.geometry("1100x780")
90
+ root.minsize(860, 620)
91
+
92
+ BG_MAIN = "#020617"
93
+ BG_CARD = "#111827"
94
+ ACCENT = "#98c220"
95
+ ACCENT_H = "#b4de34"
96
+ FG_TEXT = "#e5e7eb"
97
+ FG_MUTED = "#9ca3af"
98
+ BORDER = "#1f2937"
99
+
100
+ LIST_BG = "#f9fafb"
101
+ LIST_FG = "#111827"
102
+ LIST_SEL_BG = "#d1fae5"
103
+ LIST_SEL_FG = "#065f46"
104
+
105
+ style = ttk.Style()
106
+ if "clam" in style.theme_names():
107
+ style.theme_use("clam")
108
+
109
+ BASE_FONT = ("Segoe UI", 10)
110
+ INFO_FONT = ("Consolas", 9)
111
+ LOG_FONT = ("Consolas", 9)
112
+
113
+ root.option_add("*Font", BASE_FONT)
114
+ root.configure(bg=BG_MAIN)
115
+
116
+ style.configure("TFrame", background=BG_MAIN)
117
+ style.configure("Card.TFrame", background=BG_CARD)
118
+
119
+ style.configure(
120
+ "Treeview",
121
+ background=LIST_BG,
122
+ fieldbackground=LIST_BG,
123
+ foreground=LIST_FG,
124
+ rowheight=24,
125
+ borderwidth=0,
126
+ highlightthickness=0,
127
+ )
128
+ style.map(
129
+ "Treeview",
130
+ background=[("selected", LIST_SEL_BG)],
131
+ foreground=[("selected", LIST_SEL_FG)],
132
+ )
133
+
134
+ style.configure(
135
+ "Accent.TButton",
136
+ background="#111827",
137
+ foreground=FG_TEXT,
138
+ padding=(10, 4),
139
+ relief="flat",
140
+ borderwidth=1,
141
+ )
142
+ style.map(
143
+ "Accent.TButton",
144
+ background=[
145
+ ("active", "#1f2937"),
146
+ ("pressed", "#111827"),
147
+ ("disabled", "#111827"),
148
+ ],
149
+ foreground=[
150
+ ("disabled", FG_MUTED),
151
+ ],
152
+ )
153
+
154
+ style.configure(
155
+ "Primary.TButton",
156
+ background=ACCENT,
157
+ foreground="#030712",
158
+ padding=(12, 5),
159
+ relief="flat",
160
+ borderwidth=0,
161
+ font=("Segoe UI", 10, "bold"),
162
+ )
163
+ style.map(
164
+ "Primary.TButton",
165
+ background=[
166
+ ("active", ACCENT_H),
167
+ ("pressed", ACCENT),
168
+ ("disabled", "#4b5563"),
169
+ ],
170
+ foreground=[
171
+ ("disabled", "#111827"),
172
+ ],
173
+ )
174
+
175
+ style.configure(
176
+ "Accent.Horizontal.TProgressbar",
177
+ troughcolor=BG_MAIN,
178
+ bordercolor=BG_MAIN,
179
+ background=ACCENT,
180
+ lightcolor=ACCENT,
181
+ darkcolor=ACCENT,
182
+ )
183
+
184
+ root.update_idletasks()
185
+ w = root.winfo_width()
186
+ h = root.winfo_height()
187
+ x = (root.winfo_screenwidth() // 2) - (w // 2)
188
+ y = (root.winfo_screenheight() // 2) - (h // 2)
189
+ root.geometry(f"{w}x{h}+{x}+{y}")
190
+
191
+
192
+ # ============================================================
193
+ # LAYOUT
194
+ # ============================================================
195
+
196
+ main_paned = ttk.Panedwindow(root, orient=tk.VERTICAL)
197
+ main_paned.pack(fill=tk.BOTH, expand=True, padx=10, pady=(8, 4))
198
+
199
+ info_outer = tk.Frame(
200
+ main_paned,
201
+ bg=BG_MAIN,
202
+ )
203
+ info_frame = tk.Frame(
204
+ info_outer,
205
+ bg=BG_CARD,
206
+ highlightbackground=BORDER,
207
+ highlightthickness=1,
208
+ bd=0,
209
+ )
210
+ info_frame.pack(fill=tk.BOTH, expand=True)
211
+ main_paned.add(info_outer, weight=1)
212
+
213
+ info_label = tk.Label(
214
+ info_frame,
215
+ text="Dataset Information",
216
+ font=("Segoe UI", 11, "bold"),
217
+ bg=BG_CARD,
218
+ fg=FG_TEXT,
219
+ )
220
+ info_label.pack(anchor="w", pady=(8, 4), padx=8)
221
+
222
+ info_text = tk.Text(
223
+ info_frame,
224
+ height=7,
225
+ wrap="word",
226
+ font=INFO_FONT,
227
+ bg=BG_MAIN,
228
+ fg=FG_TEXT,
229
+ bd=0,
230
+ relief="flat",
231
+ )
232
+ info_text.pack(fill=tk.BOTH, expand=True, padx=8, pady=(0, 8))
233
+
234
+
235
+ # ============================================================
236
+ # FILES AREA (filters + two-column panel)
237
+ # ============================================================
238
+
239
+ files_outer = tk.Frame(root, bg=BG_MAIN)
240
+ files_card = tk.Frame(
241
+ files_outer,
242
+ bg=BG_CARD,
243
+ highlightbackground=BORDER,
244
+ highlightthickness=1,
245
+ bd=0,
246
+ )
247
+ files_card.pack(fill=tk.BOTH, expand=True)
248
+ main_paned.add(files_outer, weight=3)
249
+
250
+ tree_frame = ttk.Frame(files_card, style="Card.TFrame")
251
+ tree_frame.pack(fill=tk.BOTH, expand=True, padx=6, pady=6)
252
+
253
+ filter_bar = ttk.Frame(tree_frame, padding=(4, 4), style="Card.TFrame")
254
+ filter_bar.pack(side=tk.TOP, fill=tk.X)
255
+
256
+ filter_title = ttk.Label(
257
+ filter_bar,
258
+ text="Filters",
259
+ font=("Segoe UI", 10, "bold"),
260
+ background=BG_CARD,
261
+ foreground=FG_TEXT,
262
+ )
263
+ filter_title.grid(row=0, column=0, padx=(0, 8), sticky="w")
264
+
265
+ domain_var = tk.StringVar(value="All")
266
+ year_var = tk.StringVar(value="All")
267
+ modality_var = tk.StringVar(value="All")
268
+ datatype_var = tk.StringVar(value="All")
269
+
270
+ domain_label = ttk.Label(filter_bar, text="Domain", background=BG_CARD, foreground=FG_MUTED)
271
+ year_label = ttk.Label(filter_bar, text="Year", background=BG_CARD, foreground=FG_MUTED)
272
+ modality_label = ttk.Label(filter_bar, text="Modality", background=BG_CARD, foreground=FG_MUTED)
273
+ datatype_label = ttk.Label(filter_bar, text="Data type", background=BG_CARD, foreground=FG_MUTED)
274
+
275
+ domain_cb = ttk.Combobox(filter_bar, textvariable=domain_var, state="readonly", width=12, values=["All"])
276
+ year_cb = ttk.Combobox(filter_bar, textvariable=year_var, state="readonly", width=8, values=["All"])
277
+ modality_cb = ttk.Combobox(filter_bar, textvariable=modality_var, state="readonly", width=14, values=["All"])
278
+ datatype_cb = ttk.Combobox(filter_bar, textvariable=datatype_var, state="readonly", width=10, values=["All"])
279
+
280
+ domain_label.grid(row=0, column=1, sticky="w", padx=(0, 2))
281
+ domain_cb.grid(row=0, column=2, padx=(2, 8))
282
+
283
+ year_label.grid(row=0, column=3, sticky="w", padx=(0, 2))
284
+ year_cb.grid(row=0, column=4, padx=(2, 8))
285
+
286
+ modality_label.grid(row=0, column=5, sticky="w", padx=(0, 2))
287
+ modality_cb.grid(row=0, column=6, padx=(2, 8))
288
+
289
+ datatype_label.grid(row=0, column=7, sticky="w", padx=(0, 2))
290
+ datatype_cb.grid(row=0, column=8, padx=(2, 8))
291
+
292
+
293
+ def reset_filters():
294
+ domain_var.set("All")
295
+ year_var.set("All")
296
+ modality_var.set("All")
297
+ datatype_var.set("All")
298
+ apply_filters()
299
+
300
+
301
+ def show_metadata_only():
302
+ domain_var.set("GLOBAL")
303
+ year_var.set("All")
304
+ modality_var.set("GLOBAL")
305
+ datatype_var.set("MTD")
306
+ apply_filters()
307
+
308
+
309
+ metadata_btn = ttk.Button(filter_bar, text="Metadata", style="Accent.TButton", command=show_metadata_only)
310
+ metadata_btn.grid(row=0, column=9, padx=(4, 0), sticky="w")
311
+
312
+ reset_btn = ttk.Button(filter_bar, text="Reset", style="Accent.TButton", command=reset_filters)
313
+ reset_btn.grid(row=0, column=10, padx=(4, 0), sticky="w")
314
+
315
+ for col in range(11):
316
+ filter_bar.columnconfigure(col, weight=0)
317
+
318
+ files_split = ttk.Panedwindow(tree_frame, orient=tk.HORIZONTAL)
319
+ files_split.pack(side=tk.TOP, fill=tk.BOTH, expand=True, pady=(4, 0))
320
+
321
+ left_files_frame = ttk.Frame(files_split, padding=(0, 4), style="Card.TFrame")
322
+ files_split.add(left_files_frame, weight=3)
323
+
324
+ left_label = ttk.Label(
325
+ left_files_frame,
326
+ text="Available files",
327
+ font=("Segoe UI", 10, "bold"),
328
+ background=BG_CARD,
329
+ foreground=FG_TEXT,
330
+ )
331
+ left_label.pack(anchor="w", padx=(2, 0))
332
+
333
+ left_container = ttk.Frame(left_files_frame, style="Card.TFrame")
334
+ left_container.pack(fill=tk.BOTH, expand=True, pady=(2, 0))
335
+
336
+ left_scroll = ttk.Scrollbar(left_container, orient=tk.VERTICAL)
337
+ left_scroll.pack(side=tk.RIGHT, fill=tk.Y)
338
+
339
+ file_tree = ttk.Treeview(left_container, yscrollcommand=left_scroll.set, selectmode="none")
340
+ file_tree.pack(fill=tk.BOTH, expand=True)
341
+ left_scroll.config(command=file_tree.yview)
342
+
343
+ file_tree["columns"] = ()
344
+ file_tree.column("#0", anchor=tk.W, stretch=True)
345
+ file_tree["show"] = "tree"
346
+
347
+ right_selected_frame = ttk.Frame(files_split, padding=(8, 4), style="Card.TFrame")
348
+ files_split.add(right_selected_frame, weight=2)
349
+
350
+ right_title_bar = tk.Frame(right_selected_frame, bg=BG_CARD)
351
+ right_title_bar.pack(fill=tk.X, pady=(0, 2))
352
+
353
+ right_label = tk.Label(
354
+ right_title_bar,
355
+ text="Selected files",
356
+ font=("Segoe UI", 10, "bold"),
357
+ bg=BG_CARD,
358
+ fg=FG_TEXT,
359
+ )
360
+ right_label.pack(side=tk.LEFT)
361
+
362
+ badge_label = tk.Label(
363
+ right_title_bar,
364
+ text="[ 0 file — 0 ]",
365
+ font=("Segoe UI", 9, "bold"),
366
+ bg=ACCENT,
367
+ fg="#111827",
368
+ padx=8,
369
+ pady=2,
370
+ )
371
+ badge_label.pack(side=tk.RIGHT, padx=4)
372
+
373
+
374
+ selected_container = ttk.Frame(right_selected_frame, style="Card.TFrame")
375
+ selected_container.pack(fill=tk.BOTH, expand=True, pady=(2, 0))
376
+
377
+ selected_scroll = ttk.Scrollbar(selected_container, orient=tk.VERTICAL)
378
+ selected_scroll.pack(side=tk.RIGHT, fill=tk.Y)
379
+
380
+ selected_listbox = tk.Listbox(
381
+ selected_container,
382
+ height=10,
383
+ font=LOG_FONT,
384
+ bg=LIST_BG,
385
+ fg=LIST_FG,
386
+ bd=0,
387
+ relief="flat",
388
+ selectbackground=LIST_SEL_BG,
389
+ selectforeground=LIST_SEL_FG,
390
+ )
391
+ selected_listbox.pack(fill=tk.BOTH, expand=True)
392
+ selected_listbox.config(yscrollcommand=selected_scroll.set)
393
+ selected_scroll.config(command=selected_listbox.yview)
394
+
395
+ def update_selected_panel():
396
+ """Refresh the right-hand 'selected files' list and the badge."""
397
+ selected_listbox.delete(0, tk.END)
398
+
399
+ total_bytes = 0
400
+ selected_items = [m for m in dataset_files if checkbox_state.get(m["path"], False)]
401
+
402
+ for meta in sorted(selected_items, key=lambda m: m["name"]):
403
+ selected_listbox.insert(tk.END, meta["name"])
404
+ if meta.get("size_bytes") is not None:
405
+ total_bytes += meta["size_bytes"]
406
+
407
+ count = len(selected_items)
408
+ size_str = human_bytes(total_bytes) if total_bytes > 0 else "0"
409
+
410
+ badge_label.config(
411
+ text=f"[ {count} file{'s' if count != 1 else ''} — {size_str} ]"
412
+ )
413
+
414
+
415
+ # ============================================================
416
+ # BOTTOM CONTROLS / LOGS
417
+ # ============================================================
418
+
419
+ bottom_outer = tk.Frame(root, bg=BG_MAIN)
420
+ bottom_card = tk.Frame(
421
+ bottom_outer,
422
+ bg=BG_CARD,
423
+ highlightbackground=BORDER,
424
+ highlightthickness=1,
425
+ bd=0,
426
+ )
427
+ bottom_card.pack(fill=tk.X, padx=10, pady=(2, 0))
428
+ bottom_outer.pack(fill=tk.X)
429
+
430
+ bottom_frame = ttk.Frame(bottom_card, padding=(10, 8), style="Card.TFrame")
431
+ bottom_frame.pack(fill=tk.X)
432
+
433
+ path_label = ttk.Label(bottom_frame, text="Download to:", background=BG_CARD, foreground=FG_MUTED)
434
+ path_label.grid(row=0, column=0, sticky="w")
435
+
436
+ download_path = ttk.Entry(bottom_frame)
437
+ download_path.insert(0, DEFAULT_DOWNLOAD_DIR)
438
+ download_path.grid(row=0, column=1, sticky="ew", padx=(6, 6))
439
+
440
+
441
+ def choose_folder():
442
+ folder = filedialog.askdirectory()
443
+ if folder:
444
+ download_path.delete(0, tk.END)
445
+ download_path.insert(0, folder)
446
+
447
+
448
+ browse_btn = ttk.Button(bottom_frame, text="📂 Browse…", style="Accent.TButton", command=choose_folder)
449
+ browse_btn.grid(row=0, column=2, sticky="ew")
450
+
451
+ btn_frame = ttk.Frame(bottom_frame, style="Card.TFrame")
452
+ btn_frame.grid(row=0, column=3, sticky="e", padx=(10, 0))
453
+
454
+ select_all_btn = ttk.Button(btn_frame, text="Select All", style="Accent.TButton")
455
+ unselect_all_btn = ttk.Button(btn_frame, text="Unselect All", style="Accent.TButton")
456
+ download_btn = ttk.Button(btn_frame, text="⬇ Download", style="Primary.TButton")
457
+ stop_btn = ttk.Button(btn_frame, text="⛔ Stop", style="Accent.TButton", state="disabled")
458
+
459
+ select_all_btn.pack(side=tk.LEFT, padx=(0, 4))
460
+ unselect_all_btn.pack(side=tk.LEFT, padx=(0, 4))
461
+ download_btn.pack(side=tk.LEFT, padx=(0, 4))
462
+ stop_btn.pack(side=tk.LEFT)
463
+
464
+ bottom_frame.columnconfigure(1, weight=1)
465
+
466
+ status_outer = tk.Frame(root, bg=BG_MAIN)
467
+ status_card = tk.Frame(
468
+ status_outer,
469
+ bg=BG_CARD,
470
+ highlightbackground=BORDER,
471
+ highlightthickness=1,
472
+ bd=0,
473
+ )
474
+ status_card.pack(fill=tk.X, padx=10, pady=(4, 0))
475
+ status_outer.pack(fill=tk.X)
476
+
477
+ status_frame = ttk.Frame(status_card, padding=(10, 4), style="Card.TFrame")
478
+ status_frame.pack(fill=tk.X)
479
+
480
+ status_label = ttk.Label(status_frame, text="Ready.", background=BG_CARD, foreground=FG_MUTED)
481
+ status_label.pack(side=tk.LEFT)
482
+
483
+ progress = ttk.Progressbar(status_frame, mode="indeterminate", length=220, style="Accent.Horizontal.TProgressbar")
484
+ progress.pack(side=tk.RIGHT)
485
+
486
+ log_outer = tk.Frame(root, bg=BG_MAIN)
487
+ log_card = tk.Frame(
488
+ log_outer,
489
+ bg=BG_CARD,
490
+ highlightbackground=BORDER,
491
+ highlightthickness=1,
492
+ bd=0,
493
+ )
494
+ log_card.pack(fill=tk.BOTH, expand=True, padx=10, pady=(4, 8))
495
+ log_outer.pack(fill=tk.BOTH, expand=True)
496
+
497
+ log_frame = ttk.Frame(log_card, padding=4, style="Card.TFrame")
498
+ log_frame.pack(fill=tk.BOTH, expand=True)
499
+
500
+ log_scroll = ttk.Scrollbar(log_frame, orient=tk.VERTICAL)
501
+ log_scroll.pack(side=tk.RIGHT, fill=tk.Y)
502
+
503
+ log_text = tk.Text(
504
+ log_frame,
505
+ height=10,
506
+ wrap="none",
507
+ font=LOG_FONT,
508
+ bg=BG_MAIN,
509
+ fg=FG_TEXT,
510
+ bd=0,
511
+ relief="flat",
512
+ )
513
+ log_text.pack(fill=tk.BOTH, expand=True)
514
+ log_text.config(state="disabled")
515
+ log_scroll.config(command=log_text.yview)
516
+ log_text.config(yscrollcommand=log_scroll.set)
517
+
518
+
519
+ def append_log(line: str):
520
+ log_text.config(state="normal")
521
+ log_text.insert("end", line + "\n")
522
+ log_text.see("end")
523
+ log_text.config(state="disabled")
524
+
525
+
526
+ def clear_log():
527
+ log_text.config(state="normal")
528
+ log_text.delete("1.0", "end")
529
+ log_text.config(state="disabled")
530
+
531
+
532
+ # ============================================================
533
+ # TREE / CHECKBOX LOGIC
534
+ # ============================================================
535
+
536
+ def refresh_checkboxes():
537
+ """Update tree item text based on checkbox_state."""
538
+ for iid in file_tree.get_children(""):
539
+ checked = checkbox_state.get(iid, False)
540
+ meta = next((f for f in dataset_files if f["path"] == iid), None)
541
+ if meta is None:
542
+ continue
543
+ prefix = "☑ " if checked else "☐ "
544
+ file_tree.item(iid, text=f"{prefix}{meta['name']}")
545
+ update_selected_panel()
546
+
547
+
548
+ def apply_filters(*_):
549
+ """Rebuild the left tree based on current filter values."""
550
+ file_tree.delete(*file_tree.get_children(""))
551
+
552
+ d_filter = domain_var.get()
553
+ y_filter = year_var.get()
554
+ m_filter = modality_var.get()
555
+ t_filter = datatype_var.get()
556
+
557
+ for meta in dataset_files:
558
+ if d_filter != "All" and meta["domain"] != d_filter:
559
+ continue
560
+ if y_filter != "All" and meta["year"] != y_filter:
561
+ continue
562
+ if m_filter != "All" and meta["modality"] != m_filter:
563
+ continue
564
+ if t_filter != "All" and meta["datatype"] != t_filter:
565
+ continue
566
+
567
+ path = meta["path"]
568
+ if path not in checkbox_state:
569
+ checkbox_state[path] = False
570
+
571
+ checked = checkbox_state[path]
572
+ prefix = "☑ " if checked else "☐ "
573
+ file_tree.insert("", "end", iid=path, text=f"{prefix}{meta['name']}")
574
+
575
+ update_selected_panel()
576
+
577
+
578
+ def toggle_all(state: bool):
579
+ """Select/unselect only currently visible (filtered) files."""
580
+ for iid in file_tree.get_children(""):
581
+ checkbox_state[iid] = state
582
+ refresh_checkboxes()
583
+ status_label.config(text=f"{'Selected' if state else 'Unselected'} filtered files.")
584
+
585
+
586
+ def on_tree_click(event):
587
+ region = file_tree.identify("region", event.x, event.y)
588
+ column = file_tree.identify_column(event.x)
589
+ row = file_tree.identify_row(event.y)
590
+
591
+ if region == "tree" and column == "#0" and row:
592
+ bbox = file_tree.bbox(row)
593
+ if bbox and event.x > bbox[0] + 20:
594
+ current = checkbox_state.get(row, False)
595
+ checkbox_state[row] = not current
596
+ refresh_checkboxes()
597
+
598
+
599
+ file_tree.bind("<Button-1>", on_tree_click)
600
+
601
+
602
+ # ============================================================
603
+ # DATASET LOAD
604
+ # ============================================================
605
+
606
+ def load_dataset():
607
+ global file_sizes, dataset_files
608
+ info_text.delete(1.0, tk.END)
609
+ file_tree.delete(*file_tree.get_children(""))
610
+ checkbox_state.clear()
611
+ file_sizes = {}
612
+ dataset_files = []
613
+
614
+ status_label.config(text="Contacting Hugging Face API…")
615
+ root.update_idletasks()
616
+
617
+ try:
618
+ info = api.dataset_info(DATASET_ID, files_metadata=True)
619
+ except Exception as e:
620
+ messagebox.showerror("Error", f"Failed to load: {e}")
621
+ status_label.config(text="Failed to load dataset.")
622
+ return
623
+
624
+ created = info.created_at.strftime("%Y-%m-%d %H:%M") if info.created_at else "N/A"
625
+ modified = info.lastModified.strftime("%Y-%m-%d %H:%M") if info.lastModified else "N/A"
626
+ size = getattr(info, "usedStorage", 0) or 0
627
+ size_str = f"{size/1e9:.2f} GB" if size > 1e9 else f"{size/1e6:.2f} MB"
628
+
629
+ meta_info = f"""
630
+ ID : {info.id}
631
+ Author : {info.author or 'Unknown'}
632
+ Created : {created}
633
+ Modified : {modified}
634
+ Size : {size_str}
635
+ Downloads : {info.downloads or 0}
636
+ Likes : {info.likes or 0}
637
+ Tags : {', '.join(info.tags or [])}
638
+ Tasks : {', '.join((info.cardData or {}).get('task_categories', []))}
639
+ """.rstrip()
640
+
641
+ info_text.insert(tk.END, meta_info)
642
+
643
+ for s in info.siblings:
644
+ path = s.rfilename
645
+
646
+ if path.endswith("GLOBAL_ALL_MTD.zip"):
647
+ size_attr = getattr(s, "size", None)
648
+ m = {
649
+ "path": path,
650
+ "name": os.path.basename(path),
651
+ "domain": "GLOBAL",
652
+ "year": None,
653
+ "modality": "GLOBAL",
654
+ "datatype": "MTD",
655
+ "size_bytes": size_attr,
656
+ }
657
+ dataset_files.append(m)
658
+ if size_attr is not None:
659
+ file_sizes[path] = size_attr
660
+ continue
661
+
662
+ if not path.startswith("data/"):
663
+ continue
664
+ if not path.lower().endswith(".zip"):
665
+ continue
666
+
667
+ m = parse_zip_metadata(path)
668
+ if m is None:
669
+ continue
670
+
671
+ size_attr = getattr(s, "size", None)
672
+ m["size_bytes"] = size_attr
673
+ dataset_files.append(m)
674
+
675
+ if size_attr is not None:
676
+ file_sizes[path] = size_attr
677
+
678
+ domains = sorted({f["domain"] for f in dataset_files if f["domain"]})
679
+ years = sorted({f["year"] for f in dataset_files if f["year"]})
680
+ modalities = sorted({f["modality"] for f in dataset_files if f["modality"]})
681
+ datatypes = sorted({f["datatype"] for f in dataset_files if f["datatype"]})
682
+
683
+ domain_cb["values"] = ["All"] + domains
684
+ year_cb["values"] = ["All"] + years
685
+ modality_cb["values"] = ["All"] + modalities
686
+ datatype_cb["values"] = ["All"] + datatypes
687
+
688
+ domain_var.set("All")
689
+ year_var.set("All")
690
+ modality_var.set("All")
691
+ datatype_var.set("All")
692
+
693
+ apply_filters()
694
+
695
+ status_label.config(text=f"Loaded {len(dataset_files)} ZIP files from FLAIR-HUB (tiles + metadata).")
696
+ append_log("=== Dataset loaded: FLAIR-HUB (data/*.zip + GLOBAL_ALL_MTD.zip) ===")
697
+
698
+
699
+ # ============================================================
700
+ # DOWNLOAD LOGIC
701
+ # ============================================================
702
+
703
+ def stop_download():
704
+ stop_flag.set()
705
+ status_label.config(text="⏹ Stop requested. Finishing current file…")
706
+ stop_btn.config(state="disabled")
707
+ append_log("⛔ Stop requested by user.")
708
+
709
+
710
+ def download_selected():
711
+ folder = download_path.get().strip()
712
+ os.makedirs(folder, exist_ok=True)
713
+
714
+ files_to_download = [
715
+ iid for iid in file_tree.get_children("")
716
+ if checkbox_state.get(iid, False)
717
+ ]
718
+
719
+ if not files_to_download:
720
+ messagebox.showinfo("No Selection", "No files selected (in current filter).")
721
+ return
722
+
723
+ total = len(files_to_download)
724
+ stop_flag.clear()
725
+
726
+ download_btn.config(state="disabled")
727
+ select_all_btn.config(state="disabled")
728
+ unselect_all_btn.config(state="disabled")
729
+ browse_btn.config(state="disabled")
730
+ stop_btn.config(state="normal")
731
+
732
+ progress["mode"] = "indeterminate"
733
+ progress.start(15)
734
+ status_label.config(text=f"🚀 Launching download: {total} file(s)…")
735
+ append_log(f"=== Download session started: {total} file(s) ===")
736
+
737
+ def worker():
738
+ t_session = time.time()
739
+ processed = 0
740
+
741
+ for index, fid in enumerate(files_to_download, start=1):
742
+ if stop_flag.is_set():
743
+ append_log("⛔ Download stopped by user.")
744
+ break
745
+
746
+ processed += 1
747
+
748
+ meta = next((f for f in dataset_files if f["path"] == fid), None)
749
+ name = meta["name"] if meta else fid
750
+ planned = meta["size_bytes"] if meta else None
751
+ size_info = human_bytes(planned) if planned else "unknown"
752
+
753
+ def set_label(f=name, i=index):
754
+ status_label.config(text=f"({i}/{total}) Downloading: {f}")
755
+ root.after(0, set_label)
756
+
757
+ root.after(
758
+ 0,
759
+ lambda f=name, i=index, s=size_info: append_log(
760
+ f"▶ [{i}/{total}] {f} :: size ~ {s}"
761
+ ),
762
+ )
763
+
764
+ t0 = time.time()
765
+ local_path = None
766
+ error = None
767
+
768
+ try:
769
+ local_path = hf_hub_download(
770
+ repo_id=DATASET_ID,
771
+ repo_type="dataset",
772
+ filename=fid,
773
+ local_dir=folder,
774
+ force_download=True,
775
+ )
776
+ except Exception as e:
777
+ error = e
778
+
779
+ elapsed = max(time.time() - t0, 1e-3)
780
+
781
+ size_bytes = planned
782
+ if size_bytes is None and local_path and os.path.exists(local_path):
783
+ try:
784
+ size_bytes = os.path.getsize(local_path)
785
+ except OSError:
786
+ size_bytes = None
787
+
788
+ if size_bytes and elapsed > 0:
789
+ speed_mbps = (size_bytes / elapsed) / (1024 * 1024)
790
+ else:
791
+ speed_mbps = None
792
+
793
+ if error is None:
794
+ size_h = human_bytes(size_bytes)
795
+ if speed_mbps is not None:
796
+ summary = f" ✅ Avg speed: {speed_mbps:.1f} MB/s | {size_h} in {elapsed:.1f}s"
797
+ else:
798
+ summary = f" ✅ {size_h} in {elapsed:.1f}s"
799
+ else:
800
+ summary = f" ❌ ERROR: {error}"
801
+
802
+ def push_summary(line=summary):
803
+ append_log(line)
804
+ append_log("")
805
+ root.after(0, push_summary)
806
+
807
+ def finish():
808
+ progress.stop()
809
+ stop_btn.config(state="disabled")
810
+ download_btn.config(state="normal")
811
+ select_all_btn.config(state="normal")
812
+ unselect_all_btn.config(state="normal")
813
+ browse_btn.config(state="normal")
814
+
815
+ session_time = time.time() - t_session
816
+ append_log(f"=== Download session ended :: {session_time:.1f}s ===")
817
+
818
+ if stop_flag.is_set():
819
+ status_label.config(text="⛔ Download stopped.")
820
+ messagebox.showinfo("Stopped", "Downloads were stopped.")
821
+ else:
822
+ status_label.config(text=f"✅ Completed. Processed {processed} file(s).")
823
+ messagebox.showinfo(
824
+ "Download Complete",
825
+ f"Download finished.\nProcessed {processed} file(s) in {session_time:.1f}s.",
826
+ )
827
+
828
+ root.after(0, finish)
829
+
830
+ threading.Thread(target=worker, daemon=True).start()
831
+
832
+
833
+ # ============================================================
834
+ # WIRING
835
+ # ============================================================
836
+
837
+ select_all_btn.config(command=lambda: toggle_all(True))
838
+ unselect_all_btn.config(command=lambda: toggle_all(False))
839
+ download_btn.config(command=download_selected)
840
+ stop_btn.config(command=stop_download)
841
+
842
+ domain_cb.bind("<<ComboboxSelected>>", apply_filters)
843
+ year_cb.bind("<<ComboboxSelected>>", apply_filters)
844
+ modality_cb.bind("<<ComboboxSelected>>", apply_filters)
845
+ datatype_cb.bind("<<ComboboxSelected>>", apply_filters)
846
+
847
+ load_dataset()
848
+
849
+ root.mainloop()
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/imgs/flair_bandeau.jpg ADDED

Git LFS Details

  • SHA256: 63a53d16b56525c413dc58f8b480861bb456dff38c5ac9ad189851d5cc9976e7
  • Pointer size: 131 Bytes
  • Size of remote file: 220 kB
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/setup.py ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from setuptools import setup, find_packages
2
+
3
+ with open("README.md", "r", encoding="utf-8") as fh:
4
+ long_description = fh.read()
5
+
6
+ setup(
7
+ name='flair-hub',
8
+ version='1.0.0',
9
+ author='IGNF | Anatol Garioud',
10
+ author_email='flair@ign.fr',
11
+ description='Baseline and demo code for the FLAIR-HUB dataset',
12
+ long_description=long_description,
13
+ long_description_content_type='text/markdown',
14
+ url='https://github.com/IGNF/FLAIR-HUB',
15
+ license='Apache-2.0',
16
+ project_urls={
17
+ 'Bug Tracker': 'https://github.com/IGNF/FLAIR-HUB/issues'
18
+ },
19
+ classifiers=[
20
+ 'Programming Language :: Python :: 3',
21
+ 'License :: OSI Approved :: Apache Software License',
22
+ 'Operating System :: OS Independent',
23
+ 'Topic :: Scientific/Engineering :: Artificial Intelligence',
24
+ 'Topic :: Scientific/Engineering :: Image Recognition'
25
+ ],
26
+ package_dir={'': 'src'},
27
+ packages=find_packages(where='src'),
28
+ python_requires=">=3.11,<3.14",
29
+ install_requires=[
30
+ 'geopandas==1.1.2',
31
+ 'rasterio==1.4.3',
32
+ 'omegaconf',
33
+ 'jsonargparse',
34
+ 'matplotlib',
35
+ 'pandas==2.2.3',
36
+ 'scikit-image==0.25.2',
37
+ 'scikit-learn==1.6.1',
38
+ 'pillow',
39
+ 'torchmetrics==1.7.0',
40
+ 'pytorch-lightning==2.5.1',
41
+ 'segmentation-models-pytorch==0.4.0',
42
+ 'tensorboard==2.19.0',
43
+ 'pyarrow==23.0.1',
44
+ 'pyogrio==0.10.0',
45
+ 'safetensors==0.5.3'
46
+ ],
47
+ include_package_data=True,
48
+ package_data={'': ['*.yml']},
49
+ entry_points={
50
+ 'console_scripts': [
51
+ 'flairhub=flair_hub.main:main',
52
+ 'flairhub_zonal=flair_zonal_detection.main:main',
53
+ ]
54
+ }
55
+ )
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/__init__.py ADDED
File without changes
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/data/dataloader.py ADDED
@@ -0,0 +1,258 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import torch
3
+
4
+ from typing import Dict
5
+ from torch.utils.data import Dataset
6
+
7
+ from flair_hub.data.utils_data.io import read_patch
8
+ from flair_hub.data.utils_data.norm import norm
9
+ from flair_hub.data.utils_data.augmentations import apply_numpy_augmentations
10
+ from flair_hub.data.utils_data.label import reshape_label_ohe
11
+ from flair_hub.data.utils_data.elevation import calc_elevation
12
+ from flair_hub.data.utils_data.sentinel import (
13
+ reshape_sentinel,
14
+ filter_time_series,
15
+ temporal_average,
16
+ )
17
+
18
+
19
+ class flair_dataset(Dataset):
20
+ """
21
+ PyTorch Dataset for multimodal remote sensing data.
22
+ Applies normalization, temporal aggregation, cloud filtering, and augmentations.
23
+ Args:
24
+ config (Dict): Configuration dictionary containing model and modality-specific settings, such as
25
+ input channels, normalization parameters, and temporal processing options.
26
+ dict_paths (Dict): A dictionary mapping data modalities (e.g., "AERIAL_RGBI", "LABELS")
27
+ to their corresponding file paths.
28
+ use_augmentations (callable, optional): A callable function or transformation pipeline to
29
+ apply augmentations to the samples. Defaults to None.
30
+ Methods:
31
+ __len__() -> int:
32
+ Returns the number of samples in the dataset.
33
+ __getitem__(index: int) -> Dict:
34
+ Retrieves batch_elements. Applies
35
+ normalization, temporal aggregation, and augmentations (if specified).
36
+
37
+ """
38
+
39
+ def __init__(self, config: Dict, dict_paths: Dict, use_augmentations: callable = None) -> None:
40
+
41
+ self.config = config
42
+
43
+ # flair_dataset __init__
44
+ if use_augmentations is True:
45
+ self.use_augmentations = apply_numpy_augmentations
46
+ else:
47
+ self.use_augmentations = use_augmentations
48
+
49
+
50
+ # Data and label setup (same as before)
51
+ self._init_data_paths(dict_paths)
52
+ self._init_label_info(dict_paths)
53
+ self._init_normalization()
54
+ self.ref_date = config['models']['multitemp_model']['ref_date']
55
+
56
+
57
+ def _init_data_paths(self, dict_paths):
58
+ self.list_patch = {}
59
+ enabled = self.config["modalities"]["inputs"]
60
+ for mod, enabled_flag in enabled.items():
61
+ if enabled_flag and mod in dict_paths:
62
+ self.list_patch[mod] = np.array(dict_paths[mod])
63
+ if mod == "SENTINEL2_TS":
64
+ self.list_patch["SENTINEL2_MSK-SC"] = np.array(dict_paths["SENTINEL2_MSK-SC"])
65
+
66
+ self.dict_dates = {}
67
+ if "SENTINEL2_TS" in enabled:
68
+ self.dict_dates["SENTINEL2_TS"] = dict_paths.get("DATES_S2", {})
69
+ if "SENTINEL1-ASC_TS" in enabled:
70
+ self.dict_dates["SENTINEL1-ASC_TS"] = dict_paths.get("DATES_S1_ASC", {})
71
+ if "SENTINEL1-DESC_TS" in enabled:
72
+ self.dict_dates["SENTINEL1-DESC_TS"] = dict_paths.get("DATES_S1_DESC", {})
73
+
74
+ def _init_label_info(self, dict_paths):
75
+ self.tasks = {}
76
+ for task in self.config["labels"]:
77
+ label_conf = self.config["labels_configs"][task]
78
+ self.tasks[task] = {
79
+ "data_paths": np.array(dict_paths[task]),
80
+ "num_classes": len(label_conf["value_name"]),
81
+ "channels": [label_conf.get("label_channel_nomenclature", 1)]
82
+ }
83
+
84
+ def _init_normalization(self):
85
+ self.norm_type = self.config["modalities"]["normalization"]["norm_type"]
86
+ enabled_modalities = self.config["modalities"]["inputs"]
87
+ self.channels = {
88
+ mod: self.config["modalities"]["inputs_channels"].get(mod, [])
89
+ for mod, active in enabled_modalities.items() if active
90
+ }
91
+ self.normalization = {
92
+ mod: {
93
+ "mean": self.config["modalities"]["normalization"].get(f"{mod}_means", []),
94
+ "std": self.config["modalities"]["normalization"].get(f"{mod}_stds", [])
95
+ }
96
+ for mod, active in enabled_modalities.items() if active
97
+ }
98
+
99
+ def __len__(self):
100
+ for task in self.tasks.values():
101
+ if len(task["data_paths"]) > 0:
102
+ return len(task["data_paths"])
103
+ return 0
104
+
105
+ def __getitem__(self, index):
106
+ batch = {}
107
+
108
+ # Supervision
109
+ for task, info in self.tasks.items():
110
+ batch[f"ID_{task}"] = info["data_paths"][index]
111
+ area_elem = info["data_paths"][index].split("/")[-1].split("_")
112
+ area_elem = "_".join([area_elem[0], area_elem[-2], area_elem[-1].split(".")[0]])
113
+
114
+ # AERIAL_RGBI
115
+ key = "AERIAL_RGBI"
116
+ if key in self.list_patch:
117
+ data = read_patch(self.list_patch[key][index], self.channels[key])
118
+ batch[key] = norm(data,
119
+ self.norm_type,
120
+ self.normalization[key]["mean"],
121
+ self.normalization[key]["std"]
122
+ )
123
+
124
+ # AERIAL-RLT_PAN
125
+ key = "AERIAL-RLT_PAN"
126
+ if key in self.list_patch:
127
+ data = read_patch(self.list_patch[key][index], self.channels[key])
128
+ batch[key] = norm(data,
129
+ self.norm_type,
130
+ self.normalization[key]["mean"],
131
+ self.normalization[key]["std"]
132
+ )
133
+
134
+ # DEM_ELEV
135
+ key = "DEM_ELEV"
136
+ if key in self.list_patch and self.list_patch[key][index] is not None:
137
+ zdata = read_patch(self.list_patch[key][index])
138
+ if self.config["modalities"]["pre_processings"]["calc_elevation"]:
139
+ elev_data = calc_elevation(zdata)
140
+ if self.config["modalities"]["pre_processings"]["calc_elevation_stack_dsm"]:
141
+ elev_data = np.stack((zdata[0, :, :], elev_data[0]), axis=0)
142
+ batch[key] = elev_data
143
+ else:
144
+ batch[key] = zdata
145
+ batch[key] = norm(
146
+ batch[key],
147
+ self.norm_type,
148
+ self.normalization[key]["mean"],
149
+ self.normalization[key]["std"]
150
+ )
151
+
152
+ # SPOT_RGBI
153
+ key = "SPOT_RGBI"
154
+ if key in self.list_patch:
155
+ data = read_patch(self.list_patch[key][index], self.channels[key])
156
+ batch[key] = norm(data,
157
+ self.norm_type,
158
+ self.normalization[key]["mean"],
159
+ self.normalization[key]["std"]
160
+ )
161
+
162
+ # SENTINEL2_TS
163
+ key = "SENTINEL2_TS"
164
+ if key in self.list_patch:
165
+ s2 = read_patch(self.list_patch[key][index])
166
+ s2 = reshape_sentinel(s2, chunk_size=10)[:, [x - 1 for x in self.channels[key]], :, :]
167
+
168
+ s2_dates_dict = self.dict_dates[key][area_elem]
169
+ s2_dates = s2_dates_dict["dates"]
170
+ s2_dates_diff = s2_dates_dict["diff_dates"]
171
+
172
+ if self.config["modalities"]["pre_processings"]["filter_sentinel2"]:
173
+ msk = read_patch(self.list_patch["SENTINEL2_MSK-SC"][index])
174
+ msk = reshape_sentinel(msk, chunk_size=2)
175
+ idx_valid = filter_time_series(
176
+ msk,
177
+ max_cloud_value=self.config["modalities"]["pre_processings"]["filter_sentinel2_max_cloud"],
178
+ max_snow_value=self.config["modalities"]["pre_processings"]["filter_sentinel2_max_snow"],
179
+ max_fraction_covered=self.config["modalities"]["pre_processings"]["filter_sentinel2_max_frac_cover"]
180
+ )
181
+ s2 = s2[np.where(idx_valid)[0]]
182
+ s2_dates = s2_dates[np.where(idx_valid)[0]]
183
+ s2_dates_diff = s2_dates_diff[np.where(idx_valid)[0]]
184
+
185
+ if self.config["modalities"]["pre_processings"]["temporal_average_sentinel2"]:
186
+ s2, s2_dates_diff = temporal_average(
187
+ s2, s2_dates,
188
+ period=self.config["modalities"]["pre_processings"]["temporal_average_sentinel2"],
189
+ ref_date=self.ref_date
190
+ )
191
+
192
+ batch[key] = s2
193
+ batch[key.replace("_TS", "_DATES")] = s2_dates_diff
194
+
195
+ # SENTINEL1 ASC
196
+ key = "SENTINEL1-ASC_TS"
197
+ if key in self.list_patch and self.list_patch[key][index] is not None:
198
+ sentinel1asc_data = read_patch(self.list_patch[key][index])
199
+ sentinel1asc_data = reshape_sentinel(sentinel1asc_data, chunk_size=2)[:,
200
+ [x - 1 for x in self.channels[key]], :, :]
201
+
202
+ sentinel1asc_dates_dict = self.dict_dates[key][area_elem]
203
+ sentinel1asc_dates = sentinel1asc_dates_dict['dates']
204
+ sentinel1asc_dates_diff = sentinel1asc_dates_dict['diff_dates']
205
+
206
+ if self.config["modalities"]["pre_processings"]["temporal_average_sentinel1"]:
207
+ sentinel1asc_data, sentinel1asc_dates_diff = temporal_average(
208
+ sentinel1asc_data,
209
+ sentinel1asc_dates,
210
+ period=self.config["modalities"]["pre_processings"]["temporal_average_sentinel1"],
211
+ ref_date=self.ref_date
212
+ )
213
+
214
+ batch[key] = sentinel1asc_data
215
+ batch[key.replace('_TS', '_DATES')] = sentinel1asc_dates_diff
216
+
217
+ # SENTINEL1 DESC
218
+ key = "SENTINEL1-DESC_TS"
219
+ if key in self.list_patch and self.list_patch[key][index] is not None:
220
+ sentinel1desc_data = read_patch(self.list_patch[key][index])
221
+ sentinel1desc_data = reshape_sentinel(sentinel1desc_data, chunk_size=2)[:,
222
+ [x - 1 for x in self.channels[key]], :, :]
223
+
224
+ sentinel1desc_dates_dict = self.dict_dates[key][area_elem]
225
+ sentinel1desc_dates = sentinel1desc_dates_dict['dates']
226
+ sentinel1desc_dates_diff = sentinel1desc_dates_dict['diff_dates']
227
+
228
+ if self.config["modalities"]["pre_processings"]["temporal_average_sentinel1"]:
229
+ sentinel1desc_data, sentinel1desc_dates_diff = temporal_average(
230
+ sentinel1desc_data,
231
+ sentinel1desc_dates,
232
+ period=self.config["modalities"]["pre_processings"]["temporal_average_sentinel1"],
233
+ ref_date=self.ref_date
234
+ )
235
+
236
+ batch[key] = sentinel1desc_data
237
+ batch[key.replace('_TS', '_DATES')] = sentinel1desc_dates_diff
238
+
239
+ # Labels
240
+ for task, info in self.tasks.items():
241
+ label = read_patch(info["data_paths"][index], info["channels"])
242
+ batch[task] = reshape_label_ohe(label, info["num_classes"])
243
+
244
+ # Apply numpy augmentations
245
+ if callable(self.use_augmentations):
246
+ input_keys = [k for k, v in self.config["modalities"]["inputs"].items() if v]
247
+ label_keys = list(self.config["labels"])
248
+ batch = self.use_augmentations(batch, input_keys, label_keys)
249
+
250
+ # Convert to torch tensors
251
+ batch = {
252
+ k: torch.tensor(v, dtype=torch.float32)
253
+ if isinstance(v, (np.ndarray, list)) and "ID_" not in k else v
254
+ for k, v in batch.items()
255
+ }
256
+
257
+ return batch
258
+
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/data/datamodule.py ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from torch.utils.data import DataLoader
2
+ from pytorch_lightning import LightningDataModule
3
+ from typing import Dict, Optional
4
+
5
+ from flair_hub.data.dataloader import flair_dataset
6
+ from flair_hub.data.utils_data.padding import pad_collate_flair
7
+
8
+
9
+ class FlairDataModule(LightningDataModule):
10
+ def __init__(
11
+ self,
12
+ config,
13
+ dict_train: Optional[Dict] = None,
14
+ dict_val: Optional[Dict] = None,
15
+ dict_test: Optional[Dict] = None,
16
+ num_workers: int = 1,
17
+ batch_size: int = 2,
18
+ drop_last: bool = True,
19
+ use_augmentations: bool = True,
20
+ ):
21
+ """
22
+ Initialize the FlairDataModule.
23
+ Args:
24
+ config: Configuration dictionary with settings for the dataset.
25
+ dict_train (dict, optional): Dictionary containing paths for training data.
26
+ dict_val (dict, optional): Dictionary containing paths for validation data.
27
+ dict_test (dict, optional): Dictionary containing paths for test data.
28
+ num_workers (int, optional): Number of workers for data loading.
29
+ batch_size (int, optional): Batch size for data loading.
30
+ drop_last (bool, optional): Whether to drop the last batch in training and validation.
31
+ use_augmentations (bool, optional): Whether to apply augmentations.
32
+ """
33
+ super().__init__()
34
+ self.config = config
35
+ self.dict_train = dict_train
36
+ self.dict_val = dict_val
37
+ self.dict_test = dict_test
38
+ self.batch_size = batch_size
39
+ self.num_workers = num_workers
40
+ self.drop_last = drop_last
41
+ self.use_augmentations = use_augmentations
42
+ self.train_dataset = None
43
+ self.val_dataset = None
44
+ self.pred_dataset = None
45
+
46
+ def prepare_data(self):
47
+ """Prepare the data. This is a placeholder function."""
48
+ pass
49
+
50
+ def setup(self, stage: Optional[str] = None):
51
+ """
52
+ Set up the datasets for training, validation, or prediction.
53
+ Args:
54
+ stage (str, optional): The stage for setting up the data ('fit', 'validate', 'predict').
55
+ """
56
+ if stage in ("fit", "validate"):
57
+ self._setup_train_val_datasets()
58
+ elif stage == "predict":
59
+ self._setup_pred_dataset()
60
+
61
+ def _setup_train_val_datasets(self):
62
+ """Helper function to set up the training and validation datasets."""
63
+ self.train_dataset = self._create_flair_dataset(self.dict_train, self.use_augmentations)
64
+ self.val_dataset = self._create_flair_dataset(self.dict_val, use_augmentations=None)
65
+
66
+ def _setup_pred_dataset(self):
67
+ """Helper function to set up the prediction dataset."""
68
+ self.pred_dataset = self._create_flair_dataset(self.dict_test, use_augmentations=None)
69
+
70
+ def _create_flair_dataset(self, dict_paths: Dict[str, str], use_augmentations: Optional[bool]) -> "flair_dataset":
71
+ """
72
+ Helper function to create a flair dataset.
73
+ Args:
74
+ dict_paths (dict): Dictionary containing the paths to the data.
75
+ use_augmentations (bool, optional): Whether to apply augmentations.
76
+ Returns:
77
+ flair_dataset: The created dataset.
78
+ """
79
+ return flair_dataset(
80
+ self.config,
81
+ dict_paths=dict_paths,
82
+ use_augmentations=use_augmentations
83
+ )
84
+
85
+ def _create_dataloader(self, dataset, batch_size: int, shuffle: bool, drop_last: bool):
86
+ """
87
+ Helper function to create a DataLoader for a given dataset.
88
+ Args:
89
+ dataset: The dataset to be used in the DataLoader.
90
+ batch_size (int): Batch size for the DataLoader.
91
+ shuffle (bool): Whether to shuffle the dataset.
92
+ drop_last (bool): Whether to drop the last batch.
93
+ Returns:
94
+ DataLoader: The configured DataLoader.
95
+ """
96
+ return DataLoader(
97
+ dataset=dataset,
98
+ batch_size=batch_size,
99
+ shuffle=shuffle,
100
+ num_workers=self.num_workers,
101
+ drop_last=drop_last,
102
+ collate_fn=pad_collate_flair
103
+ )
104
+
105
+ def train_dataloader(self):
106
+ """
107
+ Returns the DataLoader for training.
108
+ """
109
+ return self._create_dataloader(self.train_dataset, self.batch_size, shuffle=True, drop_last=self.drop_last)
110
+
111
+ def val_dataloader(self):
112
+ """
113
+ Returns the DataLoader for validation.
114
+ """
115
+ return self._create_dataloader(self.val_dataset, self.batch_size, shuffle=False, drop_last=self.drop_last)
116
+
117
+ def predict_dataloader(self):
118
+ """
119
+ Returns the DataLoader for prediction.
120
+ """
121
+ return self._create_dataloader(self.pred_dataset, batch_size=1, shuffle=False, drop_last=False)
122
+
123
+
124
+
125
+
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/main.py ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ import sys
3
+
4
+ from pathlib import Path
5
+
6
+ from flair_hub.tasks.stages import training_stage, predict_stage
7
+ from flair_hub.tasks.module_setup import build_data_module
8
+ from flair_hub.data.utils_data.paths import get_datasets
9
+ from flair_hub.utils.messaging import start_msg, end_msg, Logger
10
+ from flair_hub.utils.config_io import setup_environment, copy_csv_and_config
11
+ from flair_hub.utils.config_display import print_recap
12
+
13
+
14
+ argParser = argparse.ArgumentParser()
15
+ argParser.add_argument("--config", "--conf", "--c", help="Path to the .yaml config file", required=True)
16
+
17
+
18
+ def main():
19
+ """
20
+ Main function to set up the training and prediction process. It reads the config file, sets up the output folder,
21
+ initiates the training and prediction stages, and tracks emissions if enabled.
22
+ """
23
+
24
+ args = argParser.parse_args()
25
+ config, out_dir = setup_environment(args)
26
+ sys.stdout = Logger(
27
+ Path(config['paths']["out_folder"], config['paths']["out_model_name"], f'flair-compute{config["paths"]["out_model_name"]}.log').as_posix())
28
+
29
+ start_msg()
30
+
31
+ # Define datasets
32
+ dict_train, dict_val, dict_test = get_datasets(config)
33
+ print_recap(config, dict_train, dict_val, dict_test)
34
+
35
+ # Copy relevant files for tracking
36
+ if config['saving']["cp_csv_and_conf_to_output"]:
37
+ copy_csv_and_config(config, out_dir, args)
38
+
39
+ # Get LightningDataModule
40
+ dm = build_data_module(config, dict_train=dict_train, dict_val=dict_val, dict_test=dict_test)
41
+
42
+ # Initialize variable for weights
43
+ trained_state_dict = None
44
+
45
+ # Training
46
+ if config['tasks']['train']:
47
+ trained_state_dict = training_stage(config, dm, out_dir)
48
+
49
+ # Inference
50
+ if config['tasks'].get('predict') or config['tasks'].get('metrics_only'):
51
+ out_dir_predict = Path(out_dir, 'results_' + config['paths']["out_model_name"])
52
+ out_dir_predict.mkdir(parents=True, exist_ok=True)
53
+ predict_stage(config, dm, out_dir_predict, trained_state_dict)
54
+ else:
55
+ print("[WARNING] Neither prediction nor metrics_only was enabled. Finishing.")
56
+
57
+ end_msg()
58
+
59
+ if __name__ == "__main__":
60
+ main()
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/tasks/stages.py ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ import os
3
+ import datetime
4
+
5
+ from datetime import timedelta
6
+ from pathlib import Path
7
+ from typing import Dict, Optional
8
+ from collections import OrderedDict
9
+ from pytorch_lightning import seed_everything
10
+
11
+ from flair_hub.writer.prediction_writer import PredictionWriter
12
+ from flair_hub.tasks.trainers import train, predict
13
+ from flair_hub.models.checkpoint import load_checkpoint
14
+ from flair_hub.tasks.module_setup import get_input_img_sizes, build_segmentation_module
15
+
16
+
17
+
18
+ def training_stage(
19
+ config: Dict,
20
+ data_module,
21
+ out_dir: Path
22
+ ) -> OrderedDict:
23
+ """
24
+ Conducts the training stage of the model: sets up the training environment, loads the model weights
25
+ from a checkpoint if available, trains the model, and logs the training information.
26
+ Args:
27
+ config (Dict): Configuration dictionary containing parameters for the task.
28
+ data_module: Data module providing training, validation, and testing data.
29
+ out_dir (Path): Path object representing the output directory for saving checkpoints and logs.
30
+ Returns:
31
+ OrderedDict: The state dictionary of the best-performing trained model.
32
+ """
33
+ start = datetime.datetime.now()
34
+
35
+ seed_everything(config['hyperparams']['seed'], workers=True)
36
+
37
+ in_img_sizes = get_input_img_sizes(config, data_module, stage='fit')
38
+
39
+ seg_module = build_segmentation_module(config, in_img_sizes, stage='train')
40
+
41
+ if config['tasks']['train_tasks']['init_weights_only_from_ckpt']:
42
+ load_checkpoint(config, seg_module, exit_on_fail=True)
43
+
44
+ ckpt_callback = train(config, data_module, seg_module, out_dir)
45
+
46
+ best_trained_state_dict = torch.load(ckpt_callback.best_model_path, map_location=torch.device('cpu'))['state_dict']
47
+
48
+ end = datetime.datetime.now()
49
+ training_time_seconds = end - start
50
+ training_time_seconds = training_time_seconds.total_seconds()
51
+
52
+ print(f"\n[Training finished in {str(timedelta(seconds=training_time_seconds))} HH:MM:SS with "
53
+ f"{config['hardware']['num_nodes']} nodes and {config['hardware']['gpus_per_node']} gpus per node]")
54
+ print(f"Model path: {os.path.join(out_dir, 'checkpoints')}\n\n")
55
+ print('-' * 40)
56
+
57
+ return best_trained_state_dict
58
+
59
+
60
+
61
+ def predict_stage(
62
+ config: Dict,
63
+ data_module,
64
+ out_dir_predict: Path,
65
+ trained_state_dict: Optional[OrderedDict] = None
66
+ ) -> None:
67
+ """
68
+ Conducts the prediction stage of the model: runs inference and optionally computes metrics.
69
+ Handles prediction, evaluation from disk, or notifies if neither task is enabled.
70
+ Args:
71
+ config (Dict): Configuration dictionary containing prediction and evaluation settings.
72
+ data_module: Data module providing data for prediction.
73
+ out_dir_predict (Path): Directory where predictions and results are saved.
74
+ trained_state_dict (Optional[OrderedDict]): Optional state dictionary of the trained model.
75
+ If not provided, model weights will be loaded from checkpoint.
76
+ Returns:
77
+ None
78
+ """
79
+
80
+ out_dir_predict = Path(out_dir_predict)
81
+
82
+ if config['tasks'].get("metrics_only", False) and not config['tasks'].get("predict", False):
83
+ print("[ ] Running in metrics-only mode: loading predictions from disk . . .")
84
+ writer = PredictionWriter(config, out_dir_predict, write_interval="batch")
85
+ writer.load_predictions_and_compute_metrics()
86
+ return
87
+
88
+ if config['tasks'].get("predict", False):
89
+ in_img_sizes = get_input_img_sizes(config, data_module, stage="predict")
90
+ seg_module = build_segmentation_module(config, in_img_sizes, stage='predict')
91
+
92
+ if config['tasks']['train']:
93
+ seg_module.load_state_dict(trained_state_dict, strict=False)
94
+ else:
95
+ load_checkpoint(config, seg_module)
96
+
97
+ print("[ ] Running inference and metrics calculation . . .")
98
+ predict(config, data_module, seg_module, out_dir_predict)
99
+
100
+ if not config['tasks'].get("predict") and not config['tasks'].get("metrics_only"):
101
+ print("[ ] Neither 'predict' nor 'metrics_only' is enabled. Finishing.")
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/tasks/trainers.py ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import sys
3
+ import torch
4
+ import torch.nn as nn
5
+
6
+ from pathlib import Path
7
+ from typing import Dict, Any
8
+ from pytorch_lightning import Trainer
9
+ from pytorch_lightning.callbacks import EarlyStopping, ModelCheckpoint
10
+ from pytorch_lightning.callbacks.progress.tqdm_progress import TQDMProgressBar
11
+ from pytorch_lightning.loggers import TensorBoardLogger
12
+
13
+ from flair_hub.writer.prediction_writer import PredictionWriter
14
+
15
+
16
+ def check_batchnorm_and_batch_size(config: Dict[str, Any], seg_module: nn.Module) -> None:
17
+ """
18
+ Check if the model contains BatchNorm layers and if the batch size is 1.
19
+ If both conditions are met, print a message and abort the script.
20
+ Args:
21
+ config (dict): Configuration dictionary containing parameters for training.
22
+ seg_module (nn.Module): Segmentation module for training.
23
+ """
24
+ batch_size = config['hyperparams']['batch_size']
25
+
26
+ for module in seg_module.modules():
27
+ if isinstance(module, (nn.BatchNorm1d, nn.BatchNorm2d, nn.BatchNorm3d)) and batch_size == 1:
28
+ print("Warning: The model contains BatchNorm layers and the batch size is set to 1.")
29
+ print("Aborting training to avoid potential issues.")
30
+ print("Please set a batch size >1 in the current model provider configuration.")
31
+ sys.exit(1)
32
+
33
+
34
+ def train(config: Dict[str, Any], data_module: Any, seg_module: nn.Module, out_dir: str) -> ModelCheckpoint:
35
+ """
36
+ Trains a model using the provided data module and segmentation module.
37
+ Args:
38
+ config (dict): Configuration dictionary containing parameters for training.
39
+ data_module: Data module for training, validation, and testing.
40
+ seg_module (nn.Module): Segmentation module for training.
41
+ out_dir (str): Output directory for saving model checkpoints.
42
+ Returns:
43
+ ModelCheckpoint: Callback for model checkpointing.
44
+ """
45
+ check_batchnorm_and_batch_size(config, seg_module)
46
+
47
+ ckpt_callback = ModelCheckpoint(
48
+ monitor=config['saving']['ckpt_monitor'],
49
+ dirpath=os.path.join(out_dir, "checkpoints"),
50
+ filename="ckpt-{epoch:02d}-{val_loss:.2f}_" + config['paths']["out_model_name"],
51
+ save_top_k=1,
52
+ mode=config['saving']['ckpt_monitor_mode'],
53
+ save_last=config['saving']['ckpt_save_also_last'],
54
+ verbose=config['saving']['ckpt_verbose'],
55
+ save_weights_only=config['saving']['ckpt_weights_only'],
56
+ )
57
+
58
+ early_stop_callback = EarlyStopping(
59
+ monitor=config['saving']['ckpt_monitor'],
60
+ min_delta=0.00,
61
+ patience=config['saving']['ckpt_earlystopping_patience'],
62
+ mode=config['saving']['ckpt_monitor_mode'],
63
+ )
64
+
65
+ prog_rate = TQDMProgressBar(refresh_rate=config['saving']["progress_rate"])
66
+
67
+ callbacks = [
68
+ ckpt_callback,
69
+ early_stop_callback,
70
+ prog_rate,
71
+ ]
72
+
73
+ logger = TensorBoardLogger(
74
+ save_dir=out_dir,
75
+ name=Path("tensorboard_logs_" + config['paths']["out_model_name"]).as_posix(),
76
+ )
77
+
78
+ loggers = [logger]
79
+
80
+ trainer = Trainer(
81
+ accelerator=config['hardware']["accelerator"],
82
+ devices=config['hardware']["gpus_per_node"],
83
+ strategy=config['hardware']["strategy"],
84
+ num_nodes=config['hardware']["num_nodes"],
85
+ max_epochs=config['hyperparams']["num_epochs"],
86
+ num_sanity_val_steps=0,
87
+ callbacks=callbacks,
88
+ logger=loggers,
89
+ enable_progress_bar=config['saving']["enable_progress_bar"],
90
+ )
91
+
92
+ if config['tasks']['train_tasks']['resume_training_from_ckpt']:
93
+ print('---------------------------------------------------------------')
94
+ print('------------- RESUMING TRAINING FROM CKPT_PATH ----------------')
95
+ print('---------------------------------------------------------------')
96
+ trainer.fit(seg_module, datamodule=data_module, ckpt_path=config['paths']['ckpt_model_path'])
97
+ else:
98
+ trainer.fit(seg_module, datamodule=data_module)
99
+
100
+ trainer.validate(seg_module, datamodule=data_module)
101
+
102
+ return ckpt_callback
103
+
104
+
105
+ def predict(config: Dict[str, Any], data_module: Any, seg_module: nn.Module, out_dir: str) -> None:
106
+ """
107
+ This function makes predictions using the provided data module and segmentation module.
108
+ Args:
109
+ config (dict): Configuration dictionary containing parameters for prediction.
110
+ data_module: Data module for training, validation, and testing.
111
+ seg_module (nn.Module): Segmentation module for prediction.
112
+ out_dir (str): Output directory for saving the predictions.
113
+ """
114
+ writer_callback = PredictionWriter(
115
+ config,
116
+ output_dir=os.path.join(out_dir),
117
+ write_interval="batch",
118
+ )
119
+
120
+ trainer = Trainer(
121
+ accelerator=config['hardware']["accelerator"],
122
+ devices=config['hardware']["gpus_per_node"],
123
+ strategy=config['hardware']["strategy"],
124
+ num_nodes=config['hardware']["num_nodes"],
125
+ callbacks=[writer_callback],
126
+ enable_progress_bar=config['saving']["enable_progress_bar"],
127
+ logger=False,
128
+ )
129
+
130
+ trainer.predict(seg_module, datamodule=data_module, return_predictions=False)
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/utils/config_io.py ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import yaml
2
+ import os
3
+ import shutil
4
+
5
+ from pathlib import Path
6
+ from typing import Dict
7
+ from pytorch_lightning.utilities.rank_zero import rank_zero_only
8
+
9
+
10
+
11
+ def read_config(path: str) -> Dict[str, dict]:
12
+ """
13
+ Reads and combines YAML configuration(s) from a file or all files in a directory.
14
+ Args:
15
+ path (str): Path to a single YAML file or a directory containing YAML files.
16
+ Returns:
17
+ Dict[str, dict]: A combined dictionary with the contents of the YAML file(s).
18
+ """
19
+ combined_config = {}
20
+
21
+ if os.path.isfile(path) and path.endswith('.yaml'):
22
+ with open(path, "r") as f:
23
+ config = yaml.safe_load(f)
24
+ if isinstance(config, dict):
25
+ combined_config.update(config)
26
+ elif os.path.isdir(path):
27
+ for file_name in os.listdir(path):
28
+ if file_name.endswith('.yaml'):
29
+ file_path = os.path.join(path, file_name)
30
+ with open(file_path, "r") as f:
31
+ config = yaml.safe_load(f)
32
+ if isinstance(config, dict):
33
+ combined_config.update(config)
34
+ else:
35
+ raise ValueError(f"Invalid path: {path}. Must be a .yaml file or a directory containing .yaml files.")
36
+
37
+ return combined_config
38
+
39
+
40
+ def setup_environment(args) -> tuple:
41
+ """
42
+ This function reads the configuration file, creates the output directory,
43
+ and sets up the logger.
44
+ Args:
45
+ args: Command-line arguments.
46
+ Returns:
47
+ tuple: Contains the config dictionary and the output directory path.
48
+ """
49
+ config = read_config(args.config)
50
+ out_dir = Path(config['paths']["out_folder"], config['paths']["out_model_name"])
51
+ out_dir.mkdir(parents=True, exist_ok=True)
52
+ return config, out_dir
53
+
54
+
55
+ @rank_zero_only
56
+ def copy_csv_and_config(config: dict, out_dir: Path, args) -> None:
57
+ """
58
+ Copy the CSV files and configuration file to the output directory.
59
+ Args:
60
+ config (dict): Configuration dictionary.
61
+ out_dir (Path): Output directory path.
62
+ args: Command-line arguments.
63
+ """
64
+ csv_copy_dir = Path(out_dir, 'used_csv_and_config')
65
+ csv_copy_dir.mkdir(parents=True, exist_ok=True)
66
+
67
+ if config["tasks"]["train"]:
68
+ shutil.copy(config["paths"]["train_csv"], csv_copy_dir)
69
+ shutil.copy(config["paths"]["val_csv"], csv_copy_dir)
70
+
71
+ if config["tasks"]["predict"]:
72
+ shutil.copy(config["paths"]["test_csv"], csv_copy_dir)
73
+
74
+ if os.path.isdir(args.config):
75
+ shutil.copytree(args.config, csv_copy_dir, dirs_exist_ok=True)
76
+ elif os.path.isfile(args.config):
77
+ shutil.copy(args.config, csv_copy_dir / Path(args.config).name)
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_hub/utils/messaging.py ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys, os
2
+ import datetime
3
+
4
+ from pytorch_lightning.utilities.rank_zero import rank_zero_only
5
+
6
+
7
+ @rank_zero_only
8
+ def start_msg():
9
+ print("""
10
+ _____ _ _ ___ ____ _ _ _ _ ____
11
+ | ___| | / \ |_ _| _ \ | | | | | | | __ )
12
+ | |_ | | / _ \ | || |_) _____| |_| | | | | _ \
13
+ | _| | |___ / ___ \ | || _ |_____| _ | |_| | |_) |
14
+ |_| |_____/_/ \_|___|_| \_\ |_| |_|\___/|____/
15
+ _______________________________________________________
16
+
17
+ #######################################################
18
+ #################### LAUNCHING #######################
19
+ """)
20
+ print(datetime.datetime.now().strftime("Starting: %Y-%m-%d %H:%M") + '\n')
21
+ print("""
22
+ [ ] Setting up Logger . . .
23
+ [ ] Creating output files . . .
24
+ [ ] Reading config files . . .
25
+ [ ] Building up datasets . . .
26
+
27
+ """
28
+ )
29
+
30
+
31
+ @rank_zero_only
32
+ def end_msg():
33
+ print("""
34
+ #######################################################
35
+ #################### FINISHED #######################
36
+ """)
37
+ print(datetime.datetime.now().strftime("Ending: %Y-%m-%d %H:%M") + '\n')
38
+
39
+
40
+
41
+ @rank_zero_only
42
+ class Logger:
43
+ """
44
+ Custom logger that mirrors stdout to both the terminal and a log file.
45
+ Useful for capturing experiment logs during training or inference.
46
+
47
+ Attributes:
48
+ terminal (TextIO): The original system stdout.
49
+ log (TextIO): The file object to write logs to.
50
+ """
51
+
52
+ def __init__(self, filename: str = 'Default.log') -> None:
53
+ """
54
+ Initialize a Logger instance that writes both to stdout and a file.
55
+
56
+ Args:
57
+ filename (str): The log file name. Will be made unique if already exists.
58
+ """
59
+ filename = self._get_unique_filename(filename)
60
+ self.terminal = sys.stdout
61
+ self.log = open(filename, 'w', encoding='utf-8')
62
+ self.encoding = self.terminal.encoding
63
+
64
+ def _get_unique_filename(self, filename: str) -> str:
65
+ """
66
+ Generate a unique filename by appending a version suffix if needed.
67
+
68
+ Args:
69
+ filename (str): The initial filename requested.
70
+
71
+ Returns:
72
+ str: A unique filename that doesn't overwrite existing files.
73
+ """
74
+ base, ext = os.path.splitext(filename)
75
+ if not os.path.exists(filename):
76
+ return filename
77
+
78
+ version = 1
79
+ while True:
80
+ new_filename = f"{base}_v{version}{ext}"
81
+ if not os.path.exists(new_filename):
82
+ return new_filename
83
+ version += 1
84
+
85
+ def write(self, message: str) -> None:
86
+ """
87
+ Write a message to both stdout and the log file.
88
+
89
+ Args:
90
+ message (str): The string message to be logged.
91
+ """
92
+ self.terminal.write(message)
93
+ self.log.write(message)
94
+
95
+ def flush(self) -> None:
96
+ """
97
+ Flush the log file stream.
98
+ """
99
+ self.log.flush()
100
+
101
+ def close(self) -> None:
102
+ """
103
+ Close the log file stream.
104
+ """
105
+ self.log.close()
106
+
107
+ def isatty(self) -> bool:
108
+ """
109
+ Indicates whether the stream is interactive. Needed for tqdm compatibility.
110
+
111
+ Returns:
112
+ bool: Always False to ensure tqdm behaves correctly.
113
+ """
114
+ return False
models/pretrained/landcover/flair-hub-rgb-swinbase-upernet/upstream/src/flair_zonal_detection/main.py ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+
3
+ from flair_zonal_detection.inference import run_inference
4
+
5
+ def main() -> None:
6
+ parser = argparse.ArgumentParser(description="Run zonal detection inference.")
7
+ parser.add_argument(
8
+ "--config", type=str, required=True, help="Path to the detection config file"
9
+ )
10
+ args = parser.parse_args()
11
+ run_inference(args.config)
12
+
13
+ if __name__ == '__main__':
14
+ main()
models/satquery_water_bundle/MODEL_CARD.md ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # SatQuery water model
2
+
3
+ SatlasPretrain Sentinel-2 multispectral Swin-v2-Base + explicit spectral fusion head, fine-tuned on hand-labeled Sen1Floods11 with event-exclusive splits and shadow hard negatives.
4
+
5
+ - Test F1: 0.8343
6
+ - Test IoU: 0.7157
7
+ - Test precision: 0.9209
8
+ - Test recall: 0.7626
9
+ - Dark-land false-positive rate: 0.0060
10
+
11
+ Limitations: optical imagery cannot see through thick cloud. Validate on the target sensor, geography, season, and water definition. Use Sentinel-1 fusion for all-weather flood mapping.
models/satquery_water_bundle/satquery_model_health.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "inference_verified": true,
3
+ "checkpoint_sha256": "e93903a19a09b98d9d8c2ac3f9973491f03359c679e8fc5cda3c9c8432b956d5",
4
+ "verified_at": "2026-09-23T19:03:38.338051+00:00",
5
+ "output_shape": [
6
+ 1,
7
+ 2,
8
+ 512,
9
+ 512
10
+ ],
11
+ "device": "cuda",
12
+ "peak_allocated_vram_bytes": 1027438080,
13
+ "class_map": [
14
+ "background",
15
+ "water"
16
+ ],
17
+ "sensor": "Sentinel-2",
18
+ "test_metrics": {
19
+ "precision": 0.9209026999318032,
20
+ "recall": 0.7625802115746306,
21
+ "f1": 0.834296767675814,
22
+ "iou": 0.7157025429297199,
23
+ "specificity": 0.9931365170414005,
24
+ "balanced_accuracy": 0.8778583643080156,
25
+ "dark_land_fpr": 0.005988187308963608
26
+ }
27
+ }
models/satquery_water_bundle/satquery_water_config.json ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format_version": 1,
3
+ "task": "surface_water_segmentation",
4
+ "architecture": "SatlasWaterNet",
5
+ "pretrained_model": "Sentinel2_SwinB_SI_MS",
6
+ "source_band_order": [
7
+ "B1",
8
+ "B2",
9
+ "B3",
10
+ "B4",
11
+ "B5",
12
+ "B6",
13
+ "B7",
14
+ "B8",
15
+ "B8A",
16
+ "B9",
17
+ "B10",
18
+ "B11",
19
+ "B12"
20
+ ],
21
+ "satlas_band_order": [
22
+ "R_from_B4",
23
+ "G_from_B3",
24
+ "B_from_B2",
25
+ "B5",
26
+ "B6",
27
+ "B7",
28
+ "B8",
29
+ "B11",
30
+ "B12"
31
+ ],
32
+ "spectral_features": [
33
+ "NDWI",
34
+ "MNDWI",
35
+ "NDVI",
36
+ "AWEI_shadow",
37
+ "visible_brightness",
38
+ "B10_cirrus"
39
+ ],
40
+ "normalization": {
41
+ "RGB": "clip(raw/3000,0,1)",
42
+ "B5_B6_B7_B8_B11_B12": "clip(raw/8160,0,1)",
43
+ "source_scale": 10000
44
+ },
45
+ "postprocess": {
46
+ "threshold": 0.5999999999999999,
47
+ "min_area_pixels": 16
48
+ },
49
+ "tile_inference": {
50
+ "tile_size": 512,
51
+ "overlap": 128,
52
+ "blend": "Hann"
53
+ },
54
+ "event_split": {
55
+ "Bolivia": "test",
56
+ "Ghana": "val",
57
+ "India": "train",
58
+ "Mekong": "train",
59
+ "Nigeria": "train",
60
+ "Pakistan": "train",
61
+ "Paraguay": "test",
62
+ "Somalia": "train",
63
+ "Spain": "val",
64
+ "Sri-Lanka": "train",
65
+ "USA": "train"
66
+ },
67
+ "seed": 20260922
68
+ }
models/satquery_water_bundle/satquery_water_metrics.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "validation": {
3
+ "precision": 0.8908810168728125,
4
+ "recall": 0.8982742527951183,
5
+ "f1": 0.8945623594863392,
6
+ "iou": 0.8092381937263012,
7
+ "specificity": 0.9885565110100245,
8
+ "balanced_accuracy": 0.9434153819025715,
9
+ "dark_land_fpr": 0.008992282683002863
10
+ },
11
+ "test": {
12
+ "precision": 0.9209026999318032,
13
+ "recall": 0.7625802115746306,
14
+ "f1": 0.834296767675814,
15
+ "iou": 0.7157025429297199,
16
+ "specificity": 0.9931365170414005,
17
+ "balanced_accuracy": 0.8778583643080156,
18
+ "dark_land_fpr": 0.005988187308963608
19
+ },
20
+ "per_test_event": {
21
+ "Bolivia": {
22
+ "precision": 0.9277822733037633,
23
+ "recall": 0.8797047029441767,
24
+ "f1": 0.9031040774941685,
25
+ "iou": 0.8233270440380979,
26
+ "specificity": 0.9870917395338736,
27
+ "balanced_accuracy": 0.9333982212390252,
28
+ "dark_land_fpr": 0.0
29
+ },
30
+ "Paraguay": {
31
+ "precision": 0.9182150207252384,
32
+ "recall": 0.7245027120250382,
33
+ "f1": 0.8099374097871166,
34
+ "iou": 0.6805838755440851,
35
+ "specificity": 0.9940910843947834,
36
+ "balanced_accuracy": 0.8592968982099108,
37
+ "dark_land_fpr": 0.0
38
+ }
39
+ }
40
+ }
models/satquery_water_bundle/satquery_water_state_dict.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e93903a19a09b98d9d8c2ac3f9973491f03359c679e8fc5cda3c9c8432b956d5
3
+ size 360362139
models/satquery_water_bundle/thresholds.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "water_probability_threshold": 0.6,
3
+ "minimum_component_area_pixels": 16,
4
+ "uncertainty_threshold": 0.3
5
+ }
models/water/s2-water-unetplusplus-efficientnet-b4/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "encoder_name": "efficientnet-b4",
3
+ "architecture": "unetplusplus",
4
+ "num_channels": 6,
5
+ "num_classes": 2,
6
+ "use_timm_model": false,
7
+ "timm_model_name": null,
8
+ "model_type": "timm_segmentation"
9
+ }
models/water/s2-water-unetplusplus-efficientnet-b4/model.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb8ae300b8b8e929fa764d72ac476a419e8eb9f121e52cce6c0c8816a48d487e
3
+ size 84032530