README.md

---
license: other
license_name: open-licence-2.0
license_link: https://www.etalab.gouv.fr/wp-content/uploads/2018/11/open-licence.pdf
pretty_name: French Land Cover from Aerospace Imagery
size_categories:
- 10B<n<100B
---

# Datset Card for FLAIR land-cover semantic segmentation

## Context & Data
<hr style='margin-top:-1em' />

The hereby FLAIR (#2) dataset is sampled countrywide and is composed of over 20 billion annotated pixels of very high resolution aerial imagery at 0.2 m spatial resolution, acquired over three years and different months (spatio-temporal domains). 
Aerial imagery patches consist of 5 channels (RVB-Near Infrared-Elevation) and have corresponding annotation (with 19 semantic classes or 13 for the baselines). 
Furthermore, to integrate broader spatial context and temporal information, high resolution Sentinel-2 satellite 1-year time series with 10 spectral band are also provided. 
More than 50,000 Sentinel-2 acquisitions with 10 m spatial resolution are available.
<br>

The dataset covers 50 spatial domains, encompassing 916 areas spanning 817 km². This dataset provides a robust foundation for advancing land cover mapping techniques.<br><br>

<style type="text/css">
.tg  {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  overflow:hidden;padding:2px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
  font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-km2t{border-color:#ffffff;font-weight:bold;text-align:left;vertical-align:top}
.tg .tg-rime{background-color:#E4DF7C;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-r3rw{background-color:#a97101;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-zv4m{border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-nto1{background-color:#000000;border-color:inherit;text-align:left;vertical-align:top}
.tg .tg-9efv{background-color:#938e7b;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-8jgo{border-color:#ffffff;text-align:center;vertical-align:top}
.tg .tg-b45e{background-color:#194A26;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-9xgv{background-color:#1553ae;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-3m6m{background-color:#f80c00;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-2e1p{background-color:#db0e9a;border-color:#ffffff;color:#db0e9a;text-align:left;vertical-align:top}
.tg .tg-l5fa{background-color:#FFF30D;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-2cns{background-color:#3DE6EB;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-jjsp{background-color:#FFF;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-2w6m{background-color:#8AB3A0;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-nla7{background-color:#6B714F;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-qg2z{background-color:#46E483;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-nv8o{background-color:#C5DC42;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-grz5{background-color:#F3A60D;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-bja1{background-color:#99F;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-69kt{background-color:#660082;border-color:#ffffff;text-align:left;vertical-align:top}
.tg .tg-r1r4{background-color:#5F0;border-color:#ffffff;text-align:left;vertical-align:top}
</style>
<table class="tg">
<thead>
  <tr>
    <th class="tg-zv4m"></th>
    <th class="tg-zv4m">Class</th>
    <th class="tg-8jgo">Freq.-train (%)</th>
    <th class="tg-8jgo">Freq.-test (%)</th>
    <th class="tg-zv4m"></th>
    <th class="tg-zv4m">Class</th>
    <th class="tg-8jgo">Freq.-train (%)</th>
    <th class="tg-8jgo">Freq.-test (%)</th>
  </tr>
</thead>
<tbody>
  <tr>
    <td class="tg-2e1p"></td>
    <td class="tg-km2t">(1) Building</td>
    <td class="tg-8jgo">8.14</td>
    <td class="tg-8jgo">3.26</td>
    <td class="tg-l5fa"></td>
    <td class="tg-km2t">(11) Agricultural Land</td>
    <td class="tg-8jgo">10.98</td>
    <td class="tg-8jgo">18.19</td>
  </tr>
  <tr>
    <td class="tg-9efv"></td>
    <td class="tg-km2t">(2) Pervious surface</td>
    <td class="tg-8jgo">8.25</td>
    <td class="tg-8jgo">3.82</td>
    <td class="tg-rime"></td>
    <td class="tg-km2t">(12) Plowed land</td>
    <td class="tg-8jgo">3.88</td>
    <td class="tg-8jgo">1.81</td>
  </tr>
  <tr>
    <td class="tg-3m6m"></td>
    <td class="tg-km2t">(3) Impervious surface</td>
    <td class="tg-8jgo">13.72</td>
    <td class="tg-8jgo">5.87</td>
    <td class="tg-2cns"></td>
    <td class="tg-km2t">(13) Swimming pool</td>
    <td class="tg-8jgo">0.01</td>
    <td class="tg-8jgo">0.02</td>
  </tr>
  <tr>
    <td class="tg-r3rw"></td>
    <td class="tg-km2t">(4) Bare soil</td>
    <td class="tg-8jgo">3.47</td>
    <td class="tg-8jgo">1.6</td>
    <td class="tg-jjsp"></td>
    <td class="tg-km2t">(14) Snow</td>
    <td class="tg-8jgo">0.15</td>
    <td class="tg-8jgo">-</td>
  </tr>
  <tr>
    <td class="tg-9xgv"></td>
    <td class="tg-km2t">(5) Water</td>
    <td class="tg-8jgo">4.88</td>
    <td class="tg-8jgo">3.17</td>
    <td class="tg-2w6m"></td>
    <td class="tg-km2t">(15) Clear cut</td>
    <td class="tg-8jgo">0.15</td>
    <td class="tg-8jgo">0.82</td>
  </tr>
  <tr>
    <td class="tg-b45e"></td>
    <td class="tg-km2t">(6) Coniferous</td>
    <td class="tg-8jgo">2.74</td>
    <td class="tg-8jgo">10.24</td>
    <td class="tg-nla7"></td>
    <td class="tg-km2t">(16) Mixed</td>
    <td class="tg-8jgo">0.05</td>
    <td class="tg-8jgo">0.12</td>
  </tr>
  <tr>
    <td class="tg-qg2z"></td>
    <td class="tg-km2t">(7) Deciduous</td>
    <td class="tg-8jgo">15.38</td>
    <td class="tg-8jgo">24.79</td>
    <td class="tg-nv8o"></td>
    <td class="tg-km2t">(17) Ligneous</td>
    <td class="tg-8jgo">0.01</td>
    <td class="tg-8jgo">-</td>
  </tr>
  <tr>
    <td class="tg-grz5"></td>
    <td class="tg-km2t">(8) Brushwood</td>
    <td class="tg-8jgo">6.95</td>
    <td class="tg-8jgo">3.81</td>
    <td class="tg-bja1"></td>
    <td class="tg-km2t">(18) Greenhouse</td>
    <td class="tg-8jgo">0.12</td>
    <td class="tg-8jgo">0.15</td>
  </tr>
  <tr>
    <td class="tg-69kt"></td>
    <td class="tg-km2t">(9) Vineyard</td>
    <td class="tg-8jgo">3.13</td>
    <td class="tg-8jgo">2.55</td>
    <td class="tg-nto1"></td>
    <td class="tg-km2t">(19) Other</td>
    <td class="tg-8jgo">0.14</td>
    <td class="tg-8jgo">0.04</td>
  </tr>
  <tr>
    <td class="tg-r1r4"></td>
    <td class="tg-km2t">(10) Herbaceous vegetation</td>
    <td class="tg-8jgo">17.84</td>
    <td class="tg-8jgo">19.76</td>
    <td class="tg-zv4m"></td>
    <td class="tg-zv4m"></td>
    <td class="tg-zv4m"></td>
    <td class="tg-zv4m"></td>
  </tr>
</tbody>
</table>

<br><br>

## Dataset Structure
<hr style='margin-top:-1em' />

The FLAIR dataset consists of 77 762 patches. Each patch includes a high-resolution aerial image (512x512) at 0.2 m, a yearly satellite image time series (40x40 by default by wider areas are provided) with a spatial resolution of 10 m 
and associated cloud and snow masks, and pixel-precise elevation and land cover annotations at 0.2 m resolution (512x512).

<p align="center"><img src="flair-patches.png" alt="" style="width:70%;max-width:600px;"/></p><br>


### Band order

<div style="display: flex;">
<div style="width: 15%;margin-right: 1;"">
Aerial
<ul>
<li>1. Red</li>
<li>2. Green</li>
<li>3. Blue</li>
<li>4. NIR</li>
<li>5. nDSM</li>
</ul>
</div>

<div style="width: 25%;">
Satellite
<ul>
<li>1. Blue (B2 490nm)</li>
<li>2. Green (B3 560nm)</li>
<li>3. Red (B4 665nm)</li>
<li>4. Red-Edge (B5 705nm)</li>
<li>5. Red-Edge2 (B6 470nm)</li>
<li>6. Red-Edge3 (B7 783nm)</li>
<li>7. NIR (B8 842nm)</li>
<li>8. NIR-Red-Edge (B8a 865nm)</li>
<li>9. SWIR (B11 1610nm)</li>
<li>10. SWIR2 (B12 2190nm)</li>
</ul>
</div>

</div>

### Annotations
Each pixel has been manually annotated by photo-interpretation of the 20 cm resolution aerial imagery, carried out by a team supervised by geography experts from the IGN. 
Movable objects like cars or boats are annotated according to their underlying cover.

### Training Splits
The dataset is made up of 50 distinct spatial domains, aligned with the administrative boundaries of the French départements. 
For our experiments, we designate 32 domains for training, 8 for validation, and reserve 10 as the official test set. 
This arrangement ensures a balanced distribution of semantic classes, radiometric attributes, bioclimatic conditions, and acquisition times across each set. 
Consequently, every split accurately reflects the landscape diversity inherent to metropolitan France. 
It is important to mention that the patches come with meta-data permitting alternative splitting schemes, for example focused on domain shifts.

Official domain split: <br/>

<div style="display: flex; flex-wrap: nowrap; align-items: center">
    <div style="flex: 40%;">
        <img src="flair-splits.png" alt="Your Image">
</div>

  <div style="flex: 60%;">
  <table border="1">
    <tr>
      <th><font color="#c7254e">TRAIN:</font></th>
      <td>D006, D007, D008, D009, D013, D016, D017, D021, D023, D030, D032, D033, D034, D035, D038, D041, D044, D046, D049, D051, D052, D055, D060, D063, D070, D072, D074, D078, D080, D081, D086, D091</td>
    </tr>
    <tr>
      <th><font color="#c7254e">VALIDATION:</font></th>
      <td>D004, D014, D029, D031, D058, D066, D067, D077</td>
    </tr>
    <tr>
      <th><font color="#c7254e">TEST:</font></th>
      <td>D015, D022, D026, D036, D061, D064, D068, D069, D071, D084</td>
    </tr>
  </table>

  </div>
</div>

<br><br>

## Baseline code 
<hr style='margin-top:-1em' />

We propose the U-T&T model, a two-branch architecture that combines spatial and temporal information from very high-resolution aerial images and high-resolution satellite images into a single output. The U-Net architecture is employed for the spatial/texture branch, using a ResNet34 backbone model pre-trained on ImageNet. For the spatio-temporal branch, 
the U-TAE architecture incorporates a Temporal self-Attention Encoder (TAE) to explore the spatial and temporal characteristics of the Sentinel-2 time series data, 
applying attention masks at different resolutions during decoding. This model allows for the fusion of learned information from both sources, 
enhancing the representation of mono-date and time series data.

U-T&T code repository &#128193; : https://github.com/IGNF/FLAIR-2-AI-Challenge <br/>

<th><font color="#c7254e"><b>IMPORTANT!</b></font></th> <b>The structure of the current dataset differs from the one that comes with the GitHub repository.</b> 
To work with the current dataset, you need to replace the <font color=‘#D7881C’><em>src/load_data.py</em></font> file with the one provided here. 
You also need to add the following content to the <font color=‘#D7881C’><em>flair-2-config.yml</em></font> file under the <em><b>data</b></em> tag: <br>

```
HF_data_path : " " # Path to unzipped HF dataset
domains_train : ["D006_2020","D007_2020","D008_2019","D009_2019","D013_2020","D016_2020","D017_2018","D021_2020","D023_2020","D030_2021","D032_2019","D033_2021","D034_2021","D035_2020","D038_2021","D041_2021","D044_2020","D046_2019","D049_2020","D051_2019","D052_2019","D055_2018","D060_2021","D063_2019","D070_2020","D072_2019","D074_2020","D078_2021","D080_2021","D081_2020","D086_2020","D091_2021"]
domains_val : ["D004_2021","D014_2020","D029_2021","D031_2019","D058_2020","D066_2021","D067_2021","D077_2021"]   
domains_test : ["D015_2020","D022_2021","D026_2020","D036_2020","D061_2020","D064_2021","D068_2021","D069_2020","D071_2020","D084_2021"]
```


<br><br>


## Reference
<hr style='margin-top:-1em' />

Please include a citation to the following article if you use the FLAIR dataset:

```
@inproceedings{garioud2023flair,
      title={FLAIR: a Country-Scale Land Cover Semantic Segmentation Dataset From Multi-Source Optical Imagery}, 
      author={Anatol Garioud and Nicolas Gonthier and Loic Landrieu and Apolline De Wit and Marion Valette and Marc Poupée and Sébastien Giordano and Boris Wattrelos},
      year={2023},
      booktitle={Advances in Neural Information Processing Systems (NeurIPS) 2023},
      doi={https://doi.org/10.48550/arXiv.2310.13336},
}
```

## Acknowledgment
<hr style='margin-top:-1em' />

This work was performed using HPC/AI resources from GENCI-IDRIS (Grant 2022-A0131013803). This work was supported by the project "Copernicus / FPCUP” of the European Union, by the French Space Agency (CNES) and by Connect by CNES.<br>


## Contact
<hr style='margin-top:-1em' />

If you have any questions, issues or feedback, you can contact us at: ai-challenge@ign.fr


## Dataset license
<hr style='margin-top:-1em' />

The "OPEN LICENCE 2.0/LICENCE OUVERTE" is a license created by the French government specifically for the purpose of facilitating the dissemination of open data by public administration.<br/>
This licence is governed by French law.<br/>
This licence has been designed to be compatible with any free licence that at least requires an acknowledgement of authorship, and specifically with the previous version of this licence as well as with the following licences: United Kingdom’s “Open Government Licence” (OGL), Creative Commons’ “Creative Commons Attribution” (CC-BY) and Open Knowledge Foundation’s “Open Data Commons Attribution” (ODC-BY).