---
license: etalab-2.0
pipeline_tag: image-segmentation
tags:
- semantic segmentation
- pytorch
- landcover
model-index:
- name: FLAIR-HUB_LC-A_RVB_swinlarge-upernet
results:
- task:
type: semantic-segmentation
dataset:
name: IGNF/FLAIR-HUB/
type: earth-observation-dataset
metrics:
- type: mIoU
value: 63.356
name: mIoU
- type: OA
value: 76.954
name: Overall Accuracy
- type: IoU
value: 83.972
name: IoU building
- type: IoU
value: 77.247
name: IoU greenhouse
- type: IoU
value: 61.59
name: IoU swimming pool
- type: IoU
value: 75.642
name: IoU impervious surface
- type: IoU
value: 57.941
name: IoU pervious surface
- type: IoU
value: 63.61
name: IoU bare soil
- type: IoU
value: 90.07
name: IoU water
- type: IoU
value: 54.777
name: IoU snow
- type: IoU
value: 53.235
name: IoU herbaceous vegetation
- type: IoU
value: 57.935
name: IoU agricultural land
- type: IoU
value: 38.391
name: IoU plowed land
- type: IoU
value: 78.814
name: IoU vineyard
- type: IoU
value: 69.909
name: IoU deciduous
- type: IoU
value: 59.468
name: IoU coniferous
- type: IoU
value: 30.173
name: IoU brushwood
library_name: pytorch
---
π FLAIR-HUB Model Collection
-
Trained on: FLAIR-HUB dataset
π
-
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
π
Model ID |
πΊοΈ
Land-cover |
πΎ
Crop-types |
π©οΈ
Aerial |
β°οΈ
Elevation |
π°οΈ
SPOT |
π°οΈ
S2 t.s. |
π°οΈ
S1 t.s. |
π©οΈ
Historical |
| LC-A |
β |
|
β |
|
|
|
|
|
| LC-D |
β |
|
β |
|
|
β |
|
|
| LC-F |
β |
|
β |
|
|
β |
β |
|
| LC-G |
β |
|
|
|
|
β |
|
|
| LC-I |
β |
|
|
|
β |
|
|
|
| LC-L |
β |
|
β |
β |
β |
β |
β |
|
| LPIS-A |
|
β |
β |
|
|
|
|
|
| LPIS-F |
|
β |
|
|
|
β |
|
|
| LPIS-I |
|
β |
|
|
β |
β |
β |
|
| LPIS-J |
|
β |
β |
|
β |
β |
β |
|
π Model: FLAIR-HUB_LC-A_RVB_swinlarge-upernet
-
Encoder: swin_large_patch4_window12_384
-
Decoder: upernet
-
Metrics:
| mIoU |
O.A. |
F-score |
Precision |
Recall |
| 63.36% |
76.95% |
76.35% |
77.04% |
76.37% |
-
Params.: 199.4
---
## General Informations
- **Contact:** flair@ign.fr
- **Code repository:** https://github.com/IGNF/FLAIR-HUB
- **Paper:** http://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
```yaml
- Model architecture: swin_large_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] β weight = 0
- Input channels:
- AERIAL_RGBI : [1,2,3]
- Input normalization (custom):
- AERIAL_RGBI:
mean: [105.66, 111.35, 102.18]
std: [52.23, 45.62, 44.30]
```
---
### Training Data
```yaml
- Train patches: 152225
- Validation patches: 38175
- Test patches: 50700
```
---
### Training Logging
---
## Metrics
| Metric | Value |
| ---------------- | ------ |
| mIoU | 63.36% |
| Overall Accuracy | 76.95% |
| F-score | 76.35% |
| Precision | 77.04% |
| Recall | 76.37% |
| Class | IoU (%) | F-score (%) | Precision (%) | Recall (%) |
| --------------------- | ------- | ----------- | ------------- | ---------- |
| building | 83.97 | 91.29 | 91.49 | 91.08 |
| greenhouse | 77.25 | 87.16 | 84.38 | 90.14 |
| swimming pool | 59.15 | 74.33 | 73.53 | 75.15 |
| impervious surface | 75.64 | 86.13 | 86.24 | 86.02 |
| pervious surface | 57.94 | 73.37 | 71.93 | 74.87 |
| bare soil | 63.61 | 77.76 | 73.29 | 82.81 |
| water | 90.07 | 94.78 | 94.50 | 95.05 |
| snow | 54.78 | 70.78 | 92.39 | 57.37 |
| herbaceous vegetation | 53.23 | 69.48 | 72.51 | 66.69 |
| agricultural land | 57.93 | 73.37 | 69.54 | 77.64 |
| plowed land | 38.39 | 55.48 | 53.90 | 57.16 |
| vineyard | 78.81 | 88.15 | 85.33 | 91.17 |
| deciduous | 69.91 | 82.29 | 81.36 | 83.24 |
| coniferous | 59.47 | 74.58 | 78.84 | 70.76 |
| brushwood | 30.17 | 46.36 | 46.41 | 46.31 |
---
## 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, SΓ©bastien 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, SΓ©bastien 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
```
