metadata

license: etalab-2.0
pipeline_tag: image-segmentation
tags:
  - semantic segmentation
  - pytorch
  - landcover
library_name: pytorch
model-index:
  - name: FLAIR-HUB_LC-A_swinlarge-upernet
    results:
      - task:
          type: semantic-segmentation
        dataset:
          name: IGNF/FLAIR-HUB/
          type: earth-observation-dataset
        metrics:
          - type: mIoU
            value: 61.868
            name: mIoU
          - type: OA
            value: 76.067
            name: Overall Accuracy
          - type: IoU
            value: 83.423
            name: IoU building
          - type: IoU
            value: 75.669
            name: IoU greenhouse
          - type: IoU
            value: 61.59
            name: IoU swimming pool
          - type: IoU
            value: 75.15
            name: IoU impervious surface
          - type: IoU
            value: 56.467
            name: IoU pervious surface
          - type: IoU
            value: 62.006
            name: IoU bare soil
          - type: IoU
            value: 88.373
            name: IoU water
          - type: IoU
            value: 51.815
            name: IoU snow
          - type: IoU
            value: 52.26
            name: IoU herbaceous vegetation
          - type: IoU
            value: 56.434
            name: IoU agricultural land
          - type: IoU
            value: 34.136
            name: IoU plowed land
          - type: IoU
            value: 77.787
            name: IoU vineyard
          - type: IoU
            value: 69.129
            name: IoU deciduous
          - type: IoU
            value: 57.445
            name: IoU coniferous
          - type: IoU
            value: 28.85
            name: IoU brushwood

🌐 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_swinsmall-upernet

Encoder: swin_small_patch4_window7_224
Decoder: upernet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
61.87%	76.07%	75.11%	76.13%	74.97%

Params.: 50.7

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_small_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	61.87%
Overall Accuracy	76.07%
F-score	75.11%
Precision	76.13%
Recall	74.97%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	83.42	90.96	91.14	90.79
greenhouse	75.67	86.15	84.00	88.42
swimming pool	59.08	74.28	73.34	75.24
impervious surface	75.15	85.81	85.80	85.82
pervious surface	56.47	72.18	71.49	72.88
bare soil	62.01	76.55	72.14	81.53
water	88.37	93.83	92.76	94.92
snow	51.82	68.26	92.98	53.93
herbaceous vegetation	52.26	68.65	72.10	65.51
agricultural land	56.43	72.15	67.57	77.40
plowed land	34.14	50.90	49.04	52.90
vineyard	77.79	87.51	85.54	89.56
deciduous	69.13	81.75	80.27	83.28
coniferous	57.44	72.97	77.92	68.61
brushwood	28.85	44.78	45.88	43.73

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