metadata

license: etalab-2.0
pipeline_tag: image-segmentation
tags:
  - semantic segmentation
  - pytorch
  - landcover
model-index:
  - name: FLAIR-HUB_LC-A_RVB_swintiny-upernet
    results:
      - task:
          type: semantic-segmentation
        dataset:
          name: IGNF/FLAIR-HUB/
          type: earth-observation-dataset
        metrics:
          - type: mIoU
            value: 62.007
            name: mIoU
          - type: OA
            value: 75.58
            name: Overall Accuracy
          - type: IoU
            value: 82.435
            name: IoU building
          - type: IoU
            value: 76.304
            name: IoU greenhouse
          - type: IoU
            value: 61.59
            name: IoU swimming pool
          - type: IoU
            value: 74.236
            name: IoU impervious surface
          - type: IoU
            value: 55.914
            name: IoU pervious surface
          - type: IoU
            value: 62.431
            name: IoU bare soil
          - type: IoU
            value: 88.028
            name: IoU water
          - type: IoU
            value: 60.68
            name: IoU snow
          - type: IoU
            value: 51.072
            name: IoU herbaceous vegetation
          - type: IoU
            value: 56.233
            name: IoU agricultural land
          - type: IoU
            value: 33.876
            name: IoU plowed land
          - type: IoU
            value: 77.533
            name: IoU vineyard
          - type: IoU
            value: 68.486
            name: IoU deciduous
          - type: IoU
            value: 55.445
            name: IoU coniferous
          - type: IoU
            value: 29.469
            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.
LC-A	✓		✓
LC-D	✓		✓			✓
LC-F	✓		✓			✓	✓
LC-G	✓					✓
LC-I	✓				✓
LC-L	✓		✓	✓	✓	✓	✓
LPIS-A		✓	✓
LPIS-F		✓				✓
LPIS-I		✓			✓	✓	✓
LPIS-J		✓	✓		✓	✓	✓

🔍 Model: FLAIR-HUB_LC-A_RVB_swintiny-upernet

Encoder: swin_tiny_patch4_window7_224
Decoder: upernet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
62.01%	75.58%	75.27%	76.11%	75.10%

Params.: 29.4

General Informations

Contact: [email protected]
Code repository: https://github.com/IGNF/FLAIR-HUB
Paper: https://arxiv.org/abs/2506.07080
Developed by: IGN
Compute infrastructure:
- software: python, pytorch-lightning
- hardware: HPC/AI resources provided by GENCI-IDRIS
License: Etalab 2.0

Training Config Hyperparameters

- Model architecture: swin_tiny_patch4_window7_224-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 : [1,2,3]
- Input normalization (custom):
    - AERIAL_RGBI:
        mean: [105.66, 111.35, 102.18]
        std:  [52.23, 45.62, 44.30]

Training Data

- Train patches: 152225
- Validation patches: 38175
- Test patches: 50700

Training Logging

Metrics

Metric	Value
mIoU	62.01%
Overall Accuracy	75.58%
F-score	75.27%
Precision	76.11%
Recall	75.10%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	82.44	90.37	90.51	90.24
greenhouse	76.30	86.56	84.06	89.21
swimming pool	57.97	73.39	73.87	72.92
impervious surface	74.24	85.21	85.38	85.05
pervious surface	55.91	71.72	69.81	73.75
bare soil	62.43	76.87	73.25	80.86
water	88.03	93.63	93.18	94.09
snow	60.68	75.53	94.83	62.76
herbaceous vegetation	51.07	67.61	71.85	63.84
agricultural land	56.23	71.99	67.08	77.67
plowed land	33.88	50.61	50.90	50.32
vineyard	77.53	87.35	83.85	91.14
deciduous	68.49	81.30	79.27	83.43
coniferous	55.44	71.34	79.03	65.01
brushwood	29.47	45.52	44.80	46.27

Inference

Aerial ROI

Inference ROI

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