metadata

license: etalab-2.0
pipeline_tag: image-segmentation
tags:
  - semantic segmentation
  - pytorch
  - landcover
library_name: pytorch
model-index:
  - name: FLAIR-HUB_LC-A_swinbase-unet
    results:
      - task:
          type: semantic-segmentation
        dataset:
          name: IGNF/FLAIR-HUB/
          type: earth-observation-dataset
        metrics:
          - type: mIoU
            value: 64.803
            name: mIoU
          - type: OA
            value: 77.93
            name: Overall Accuracy
          - type: IoU
            value: 84.7
            name: IoU building
          - type: IoU
            value: 79.029
            name: IoU greenhouse
          - type: IoU
            value: 61.59
            name: IoU swimming pool
          - type: IoU
            value: 76.228
            name: IoU impervious surface
          - type: IoU
            value: 57.509
            name: IoU pervious surface
          - type: IoU
            value: 64.232
            name: IoU bare soil
          - type: IoU
            value: 90.6
            name: IoU water
          - type: IoU
            value: 63.761
            name: IoU snow
          - type: IoU
            value: 54.897
            name: IoU herbaceous vegetation
          - type: IoU
            value: 58.304
            name: IoU agricultural land
          - type: IoU
            value: 37.635
            name: IoU plowed land
          - type: IoU
            value: 78.314
            name: IoU vineyard
          - type: IoU
            value: 72.073
            name: IoU deciduous
          - type: IoU
            value: 62.519
            name: IoU coniferous
          - type: IoU
            value: 30.084
            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_swinbase-unet

Encoder: swin_base_patch4_window12_384
Decoder: unet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
64.80%	77.93%	77.43%	78.16%	77.17%

Params.: 92.8

General Informations

Contact: [email protected]
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

- Model architecture: swin_base_patch4_window12_384-unet
- 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]
- Input normalization (custom):
    - AERIAL_RGBI:
        mean: [106.59, 105.66, 111.35]
        std:  [39.78, 52.23, 45.62]

Training Data

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

Training Logging

Metrics

Metric	Value
mIoU	64.80%
Overall Accuracy	77.93%
F-score	77.43%
Precision	78.16%
Recall	77.17%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	84.70	91.72	91.98	91.46
greenhouse	79.03	88.29	85.94	90.77
swimming pool	62.16	76.67	76.55	76.79
impervious surface	76.23	86.51	86.75	86.28
pervious surface	57.51	73.02	70.90	75.28
bare soil	64.23	78.22	74.68	82.12
water	90.60	95.07	95.95	94.20
snow	63.76	77.87	94.88	66.03
herbaceous vegetation	54.90	70.88	73.05	68.84
agricultural land	58.30	73.66	70.66	76.93
plowed land	37.64	54.69	53.87	55.53
vineyard	78.31	87.84	85.25	90.59
deciduous	72.07	83.77	81.89	85.74
coniferous	62.52	76.94	80.55	73.64
brushwood	30.08	46.25	49.53	43.39

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