--- license: etalab-2.0 pipeline_tag: image-segmentation tags: - semantic segmentation - pytorch - landcover model-index: - name: FLAIR-HUB_LC-D_swinbase-upernet results: - task: type: semantic-segmentation dataset: name: IGNF/FLAIR-HUB/ type: earth-observation-dataset metrics: - type: mIoU value: 64.69 name: mIoU - type: OA value: 77.631 name: Overall Accuracy - type: IoU value: 83.967 name: IoU building - type: IoU value: 78.902 name: IoU greenhouse - type: IoU value: 61.59 name: IoU swimming pool - type: IoU value: 75.83 name: IoU impervious surface - type: IoU value: 57.539 name: IoU pervious surface - type: IoU value: 63.025 name: IoU bare soil - type: IoU value: 90.498 name: IoU water - type: IoU value: 68.274 name: IoU snow - type: IoU value: 54.417 name: IoU herbaceous vegetation - type: IoU value: 57.48 name: IoU agricultural land - type: IoU value: 36.857 name: IoU plowed land - type: IoU value: 78.136 name: IoU vineyard - type: IoU value: 71.93 name: IoU deciduous - type: IoU value: 62.922 name: IoU coniferous - type: IoU value: 29.421 name: IoU brushwood library_name: pytorch ---

🌐 FLAIR-HUB Model Collection

Trained on: FLAIR-HUB dataset 🔗
Available modalities: Aerial images, SPOT images, Topographic info, Sentinel-2 yearly time-series, Sentinel-1 yearly time-series, Historical aerial images
Encoders: ConvNeXTV2, Swin (Tiny, Small, Base, Large)
Decoders: UNet, UPerNet
Tasks: Land-cover mapping (LC), Crop-type mapping (LPIS)
Class nomenclature: 15 classes for LC, 23 classes for LPIS

🆔 Model ID	🗺️ Land-cover	🌾 Crop-types	🛩️ Aerial	⛰️ Elevation	🛰️ SPOT	🛰️ S2 t.s.	🛰️ S1 t.s.	🛩️ Historical
LC-A	✓		✓
LC-D	✓		✓			✓
LC-F	✓		✓			✓	✓
LC-G	✓					✓
LC-I	✓				✓
LC-L	✓		✓	✓	✓	✓	✓
LPIS-A		✓	✓
LPIS-F		✓				✓
LPIS-I		✓			✓	✓	✓
LPIS-J		✓	✓		✓	✓	✓

🔍 Model: FLAIR-HUB_LC-D_swinbase-upernet

Encoder: swin_base_patch4_window12_384
Decoder: upernet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
64.69%	77.63%	77.31%	77.65%	77.26%

Params.: 93.9

--- ## General Informations - **Contact:** flair@ign.fr - **Code repository:** https://github.com/IGNF/FLAIR-HUB - **Paper:** https://arxiv.org/abs/2506.07080 - **Project Page:** https://ignf.github.io/FLAIR/FLAIR-HUB/flairhub - **Developed by:** IGN - **Compute infrastructure:** - software: python, pytorch-lightning - hardware: HPC/AI resources provided by GENCI-IDRIS - **License:** Etalab 2.0 --- ### Training Config Hyperparameters ```yaml - Model architecture: swin_base_patch4_window12_384-upernet - Optimizer: AdamW (betas=[0.9, 0.999], weight_decay=0.01) - Learning rate: 5e-5 - Scheduler: one_cycle_lr (warmup_fraction=0.2) - Epochs: 150 - Batch size: 5 - Seed: 2025 - Early stopping: patience 20, monitor val_miou (mode=max) - Class weights: - default: 1.0 - masked classes: [clear cut, ligneous, mixed, other] → weight = 0 - Input channels: - AERIAL_RGBI : [4,1,2] - SENTINEL2_TS : [1,2,3,4,5,6,7,8,9,10] - Input normalization (custom): - AERIAL_RGBI: mean: [106.59, 105.66, 111.35] std: [39.78, 52.23, 45.62] ``` --- ### Training Data ```yaml - Train patches: 152225 - Validation patches: 38175 - Test patches: 50700 ```

Classes distribution.

--- ### Training Logging

Training logging.

--- ## Metrics | Metric | Value | | ---------------- | ------ | | mIoU | 64.69% | | Overall Accuracy | 77.63% | | F-score | 77.31% | | Precision | 77.65% | | Recall | 77.26% | | Class | IoU (%) | F-score (%) | Precision (%) | Recall (%) | | --------------------- | ------- | ----------- | ------------- | ---------- | | building | 83.97 | 91.28 | 91.16 | 91.41 | | greenhouse | 78.90 | 88.21 | 84.90 | 91.78 | | swimming pool | 61.15 | 75.89 | 74.71 | 77.11 | | impervious surface | 75.83 | 86.25 | 86.76 | 85.76 | | pervious surface | 57.54 | 73.05 | 71.89 | 74.24 | | bare soil | 63.02 | 77.32 | 73.88 | 81.09 | | water | 90.50 | 95.01 | 95.89 | 94.15 | | snow | 68.27 | 81.15 | 93.18 | 71.86 | | herbaceous vegetation | 54.42 | 70.48 | 71.80 | 69.21 | | agricultural land | 57.48 | 73.00 | 70.26 | 75.97 | | plowed land | 36.86 | 53.86 | 53.55 | 54.18 | | vineyard | 78.14 | 87.73 | 85.38 | 90.20 | | deciduous | 71.93 | 83.67 | 82.34 | 85.05 | | coniferous | 62.92 | 77.24 | 80.88 | 73.92 | | brushwood | 29.42 | 45.47 | 48.18 | 43.04 | --- ## Inference

Aerial ROI

AERIAL

Inference ROI

INFERENCE

--- ## Cite **BibTeX:** ``` @article{ign2025flairhub, doi = {10.48550/arXiv.2506.07080}, url = {https://arxiv.org/abs/2506.07080}, author = {Garioud, Anatol and Giordano, Sébastien and David, Nicolas and Gonthier, Nicolas}, title = {FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mapping}, publisher = {arXiv}, year = {2025} } ``` **APA:** ``` Anatol Garioud, Sébastien Giordano, Nicolas David, Nicolas Gonthier. FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mapping. (2025). DOI: https://doi.org/10.48550/arXiv.2506.07080 ```