metadata

license: etalab-2.0
pipeline_tag: image-segmentation
tags:
  - semantic segmentation
  - pytorch
  - landcover
library_name: pytorch
model-index:
  - name: FLAIR-HUB_LC-A_convnextv2tiny-upernet
    results:
      - task:
          type: semantic-segmentation
        dataset:
          name: IGNF/FLAIR-HUB/
          type: earth-observation-dataset
        metrics:
          - type: mIoU
            value: 62.725
            name: mIoU
          - type: OA
            value: 76.434
            name: Overall Accuracy
          - type: IoU
            value: 82.565
            name: IoU building
          - type: IoU
            value: 75.292
            name: IoU greenhouse
          - type: IoU
            value: 61.59
            name: IoU swimming pool
          - type: IoU
            value: 73.768
            name: IoU impervious surface
          - type: IoU
            value: 55.136
            name: IoU pervious surface
          - type: IoU
            value: 60.195
            name: IoU bare soil
          - type: IoU
            value: 88.596
            name: IoU water
          - type: IoU
            value: 64.808
            name: IoU snow
          - type: IoU
            value: 53.2
            name: IoU herbaceous vegetation
          - type: IoU
            value: 55.828
            name: IoU agricultural land
          - type: IoU
            value: 35.343
            name: IoU plowed land
          - type: IoU
            value: 76.054
            name: IoU vineyard
          - type: IoU
            value: 70.93
            name: IoU deciduous
          - type: IoU
            value: 60.604
            name: IoU coniferous
          - type: IoU
            value: 29.504
            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_convnextv2tiny-upernet

Encoder: convnextv2_tiny
Decoder: upernet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
62.73%	76.43%	75.87%	76.22%	75.72%

Params.: 29.8

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: convnextv2_tiny-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]
- 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	62.73%
Overall Accuracy	76.43%
F-score	75.87%
Precision	76.22%
Recall	75.72%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	82.57	90.45	90.96	89.94
greenhouse	75.29	85.90	83.32	88.66
swimming pool	59.06	74.26	75.33	73.23
impervious surface	73.77	84.90	85.42	84.40
pervious surface	55.14	71.08	70.18	72.00
bare soil	60.20	75.15	72.82	77.64
water	88.60	93.95	95.09	92.85
snow	64.81	78.65	87.46	71.45
herbaceous vegetation	53.20	69.45	70.50	68.43
agricultural land	55.83	71.65	69.34	74.12
plowed land	35.34	52.23	50.89	53.63
vineyard	76.05	86.40	84.28	88.62
deciduous	70.93	82.99	81.83	84.19
coniferous	60.60	75.47	78.67	72.52
brushwood	29.50	45.56	47.15	44.09

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