IGNF/FLAIR-HUB_LC-A_IR_swintiny-upernet

🌐 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.	🛩️ Historical
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_swintiny-upernet

• Encoder: swin_tiny_patch4_window7_224
• Decoder: upernet
• Metrics:

mIoU	O.A.	F-score	Precision	Recall
62.16%	76.24%	75.33%	75.71%	75.31%

• Params.: 29.4

---

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_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 : [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	62.16%
Overall Accuracy	76.24%
F-score	75.33%
Precision	75.71%
Recall	75.31%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
building	82.37	90.33	90.46	90.20
greenhouse	72.11	83.80	79.52	88.57
swimming pool	58.68	73.96	71.85	76.19
impervious surface	74.27	85.23	85.56	84.92
pervious surface	55.75	71.59	70.49	72.73
bare soil	60.93	75.72	74.02	77.51
water	88.45	93.87	95.17	92.61
snow	64.39	78.34	91.63	68.41
herbaceous vegetation	52.59	68.93	70.88	67.09
agricultural land	55.37	71.27	67.69	75.25
plowed land	30.81	47.10	45.73	48.56
vineyard	76.44	86.65	84.87	88.50
deciduous	70.95	83.00	81.34	84.73
coniferous	60.39	75.31	80.00	71.13
brushwood	28.89	44.83	46.43	43.33

---

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