README.md

---
license: etalab-2.0
task_categories:
- image-segmentation
language:
- en
tags:
- remote-sensing
- earth-observation
- change-detection
- weak-temporal-supervision
pretty_name: b-IAILD
size_categories:
- 10K<n<100K
viewer: false
---
# b-IAILD: bi-temporal extension of IAILD

<img src="./thumbnail.png" alt="b-IAILD" width="700">

## Dataset Description

b-IAILD is a temporal extension of the IAILD (Inria Aerial Image Labeling Dataset) dataset [1] focused on building change detection in urban areas. The dataset provides bi-temporal orthoimage pairs with binary building footprint annotations, covering locations in the USA and Austria.

Project page: https://xavibou.github.io/CDviaWTS/

## Dataset Summary

- **Task**: Building change detection via weak temporal supervision
- **Coverage**: Urban areas in USA (Austin, Chicago, Kitsap) and Austria (Tyrol, Vienna)
- **Resolution**: 0.6 m/px (standardized)
- **Patch Size**: 256×256 pixels
- **Total Training Pairs**: 15,500 (3,100 per city)
- **Total Validation Pairs**: 2,500 (500 per city)

## Dataset Structure

The dataset consists of bi-temporal image pairs with the following temporal coverage:
| Location | Original IAILD (t1) | New Acquisition (t2) | Time Gap |
|----------|-------------------|---------------------|----------|
| Austin, TX | Before 2017 | 2022 | >5 years |
| Chicago, IL | Before 2017 | 2023 | >6 years |
| Kitsap, WA | Before 2017 | 2023 | >6 years |
| Tyrol, Austria | Before 2017 | 2023 | >6 years |
| Vienna, Austria | Before 2017 | 2024 | >7 years |

## Dataset Creation

### Source Data

The dataset extends the original IAILD training set by adding new orthoimage acquisitions over the same geographic locations:

- **USA data**: https://earthexplorer.usgs.gov
- **Austrian data**: 
  - Tyrol: https://data-tiris.opendata.arcgis.com
  - Vienna: https://www.wien.gv.at/ma41datenviewer/public/start.aspx

### Preprocessing

All images were standardized to a common spatial resolution:

1. **Original resolutions** (new acquisitions):
   - Vienna: 0.15 m/px
   - Tyrol: 0.20 m/px
   - Austin, Chicago, Kitsap: 0.60 m/px
   - Original IAILD: 0.30 m/px

2. **Standardized resolution**: 0.60 m/px (all images resampled)

3. **Patching**: 2500×2500 images split into 256×256 patches with 6-pixel overlap

### Annotations

Single-temporal binary building footprint masks are provided for each pair, enabling change detection via weak temporal supervision.


## References

[1] E. Maggiori et al. (2017). Can semantic labeling methods generalize to any city? The Inria Aerial Image Labeling Benchmark. In IGARSS

## Citation

If you use this dataset, please cite the following publication:
```bibtex
@article{bou2026remote,
  title={Remote Sensing Change Detection via Weak Temporal Supervision},
  author={Bou, Xavier and Vincent, Elliot and Facciolo, Gabriele and Grompone von Gioi, Rafael and Morel, Jean-Michel and Ehret, Thibaud},
  journal={arXiv preprint arXiv:2601.02126},
  year={2026}
}
```