IGNF/FLAIR-HUB_LPIS-J_swinbase-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_LPIS-J_swinbase-upernet

• Encoder: swin_base_patch4_window12_384
• Decoder: upernet
• Metrics:

mIoU	O.A.	F-score	Precision	Recall
32.35%	87.97%	43.04%	50.96%	42.60%

• Params.: 186.9
• Code: GitHub

---

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_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]
    - SPOT_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]
    - SPOT_RGBI:
        mean: [1137.03, 433.26, 508.75]
        std:  [543.11, 312.76, 284.61]

---

Training Data

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

---

Training Logging

---

Metrics

Metric	Value
mIoU	32.35%
Overall Accuracy	87.97%
F-score	43.04%
Precision	50.96%
Recall	42.60%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
grasses	52.02	68.44	73.20	64.26
wheat	57.45	72.97	64.92	83.30
barley	30.96	47.29	72.74	35.03
maize	78.30	87.83	84.77	91.11
other cereals	8.15	15.08	18.59	12.69
rice	0.00	0.00	0.00	0.00
flax/hemp/tobacco	10.61	19.19	88.66	10.76
sunflower	45.82	62.85	53.61	75.92
rapeseed	71.89	83.64	82.53	84.78
other oilseed crops	0.00	0.00	0.00	0.00
soy	33.68	50.38	62.54	42.19
other protein crops	8.93	16.39	18.95	14.44
fodder legumes	27.19	42.76	44.10	41.50
beetroots	75.31	85.91	84.14	87.77
potatoes	14.37	25.13	19.45	35.48
other arable crops	22.10	36.20	39.78	33.22
vineyard	44.55	61.64	56.82	67.36
olive groves	16.38	28.14	55.59	18.84
fruits orchards	36.57	53.55	47.56	61.27
nut orchards	6.60	12.38	19.69	9.03
other permanent crops	12.12	21.62	84.16	12.40
mixed crops	2.30	4.50	7.39	3.24
background	88.73	94.03	92.86	95.22

---

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