Update README.md

586c6e8 verified about 14 hours ago

26.4 kB

license: etalab-2.0
pipeline_tag: image-segmentation
tags:
  - semantic segmentation
  - pytorch
  - landcover
library_name: pytorch
model-index:
  - name: FLAIR-HUB_LC-F_swinbase-upernet
    results:
      - task:
          type: semantic-segmentation
        dataset:
          name: IGNF/FLAIR-HUB/
          type: earth-observation-dataset
        metrics:
          - type: mIoU
            value: 64.868
            name: mIoU
          - type: OA
            value: 77.685
            name: Overall Accuracy
          - type: IoU
            value: 84.002
            name: IoU building
          - type: IoU
            value: 79.283
            name: IoU greenhouse
          - type: IoU
            value: 61.59
            name: IoU swimming pool
          - type: IoU
            value: 75.611
            name: IoU impervious surface
          - type: IoU
            value: 57.694
            name: IoU pervious surface
          - type: IoU
            value: 63.826
            name: IoU bare soil
          - type: IoU
            value: 90.461
            name: IoU water
          - type: IoU
            value: 68.142
            name: IoU snow
          - type: IoU
            value: 54.871
            name: IoU herbaceous vegetation
          - type: IoU
            value: 56.893
            name: IoU agricultural land
          - type: IoU
            value: 37.914
            name: IoU plowed land
          - type: IoU
            value: 78.144
            name: IoU vineyard
          - type: IoU
            value: 71.731
            name: IoU deciduous
          - type: IoU
            value: 63.69
            name: IoU coniferous
          - type: IoU
            value: 29.627
            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 task		Model inputs
🆔 Model ID	🗺️ Land-cover	🌾 Crop-types	🛩️ Aerial	⛰️ Elevation	🛰️ SPOT	🛰️ S2 t.s.	🛰️ S1 t.s.	🛩️ Historical
LC-A	✓		✓
LC-B	✓		✓	✓
LC-D	✓		✓			✓
LC-F	✓		✓			✓	✓
LC-G	✓					✓
LC-I	✓				✓
LC-L	✓		✓	✓	✓	✓	✓
LPIS-A		✓	✓
LPIS-F		✓				✓
LPIS-I		✓			✓	✓	✓
LPIS-J		✓	✓		✓	✓	✓

🔍 Model: FLAIR-HUB_LC-F_swinbase-upernet

Encoder: swin_base_patch4_window12_384
Decoder: upernet
Metrics:

mIoU	O.A.	F-score	Precision	Recall
64.87%	77.69%	77.47%	77.69%	77.73%

Params.: 97.7

General Informations

Contact: flair@ign.fr
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_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]
    - SENTINEL1-ASC_TS: [1, 2]
    - SENTINEL1-DESC_TS: [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	64.87%
Overall Accuracy	77.69%
F-score	77.47%
Precision	77.69%
Recall	77.73%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	84.00	91.31	90.86	91.76
greenhouse	79.28	88.44	85.63	91.45
swimming pool	61.13	75.88	73.91	77.95
impervious surface	75.61	86.11	87.53	84.74
pervious surface	57.69	73.17	71.90	74.49
bare soil	63.83	77.92	73.15	83.35
water	90.46	94.99	95.79	94.20
snow	68.14	81.05	96.94	69.64
herbaceous vegetation	54.87	70.86	71.99	69.77
agricultural land	56.89	72.52	70.36	74.82
plowed land	37.91	54.98	50.82	59.88
vineyard	78.14	87.73	85.52	90.05
deciduous	71.73	83.54	83.31	83.77
coniferous	63.69	77.82	77.95	77.68
brushwood	29.63	45.71	49.63	42.36

Inference

Aerial ROI

Inference ROI

Cite

BibTeX:

@article{GARIOUD2026271,
title = {FLAIR-HUB: Large-scale multimodal dataset for land cover and crop mapping},
author = {Anatol Garioud and Sébastien Giordano and Nicolas David and Nicolas Gonthier},
journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
volume = {237},
pages = {271-300},
year = {2026},
issn = {0924-2716},
doi = {https://doi.org/10.1016/j.isprsjprs.2026.04.017},
url = {https://www.sciencedirect.com/science/article/pii/S0924271626001899},
}

APA:

Anatol Garioud, Sébastien Giordano, Nicolas David, Nicolas Gonthier. 
FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mapping.
ISPRS Journal of Photogrammetry and Remote Sensing, Volume 237, 2026.
DOI: https://doi.org/10.1016/j.isprsjprs.2026.04.017