IGNF/FLAIR-HUB_LPIS-I_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-I_swinbase-upernet

• Encoder: swin_base_patch4_window12_384
• Decoder: upernet
• Metrics:

mIoU	O.A.	F-score	Precision	Recall
35.76%	87.19%	46.66%	52.77%	44.45%

• Params.: 97.5

---

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-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:
    - 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):
    - 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	35.76%
Overall Accuracy	87.19%
F-score	46.66%
Precision	52.77%
Recall	44.45%

Class	IoU (%)	F-score (%)	Precision (%)	Recall (%)
grasses	47.65	64.54	68.36	61.13
wheat	65.72	79.32	76.87	81.93
barley	45.99	63.00	69.21	57.82
maize	74.46	85.36	79.16	92.61
other cereals	13.98	24.54	26.33	22.97
rice	0.00	0.00	0.00	0.00
flax/hemp/tobacco	56.98	72.59	85.52	63.06
sunflower	44.07	61.17	62.25	60.14
rapeseed	81.60	89.87	86.69	93.29
other oilseed crops	0.00	0.00	0.00	0.00
soy	51.80	68.24	75.15	62.50
other protein crops	8.65	15.93	18.03	14.26
fodder legumes	28.25	44.05	50.58	39.01
beetroots	75.18	85.83	91.19	81.07
potatoes	7.18	13.41	51.09	7.71
other arable crops	22.77	37.10	32.97	42.41
vineyard	33.02	49.64	58.03	43.37
olive groves	14.16	24.80	25.63	24.02
fruit orchards	27.82	43.53	49.41	38.90
nut orchards	29.83	45.95	68.55	34.56
other permanent crops	0.27	0.53	20.92	0.27
mixed crops	5.49	10.42	25.67	6.53
background	87.62	93.40	92.01	94.84

---

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