isp-uv-es/bright-dfc2025-taco

BRIGHT (DFC2025 Track 2)

This is a repackaging, not a new dataset. It is BRIGHT (DFC2025 Track 2) by RIKEN AIP / University of Tokyo (Chen et al.), converted to TACO. Pixel values and labels are kept as released except where the description below says otherwise. All credit belongs to the original authors: if you use it, please cite them and follow their licence.

original dataset · paper · licence: CC-BY-NC-SA-4.0

Repackaged into TACO by the Image and Signal Processing Group (ISP), Universitat de València, within the ELLIOT project.

Citation

Please cite the original work:

@article{chen2025bright,
  title   = {BRIGHT: a globally distributed multimodal building damage assessment dataset with very-high-resolution for all-weather disaster response},
  author  = {Chen, Hongruixuan and Song, Jian and Dietrich, Olivier and Broni-Bediako, Clifford and Xuan, Weihao and Wang, Junjue and Shao, Xinlei and Wei, Yimin and Xia, Junshi and Lan, Cuiling and Schindler, Konrad and Yokoya, Naoto},
  journal = {Earth System Science Data},
  volume  = {17},
  pages   = {6217--6253},
  year    = {2025},
  doi     = {10.5194/essd-17-6217-2025}
}

About the data

2890 co-registered 1024x1024 tiles over 9 disasters (bata-explosion, beirut-explosion, congo-volcano, haiti-earthquake, hawaii-wildfire, la_palma-volcano, libya-flood, morocco-earthquake, turkey-earthquake), each pairing a pre-event sub-metre optical image with a post-event sub-metre SAR image of the same ground and a four-level building damage mask (background, intact, damaged, destroyed).

2,890 samples · splits: test 660 · train 1,583 · validation 647 · tasks: semantic-segmentation

Packaged as TACO v3.

Full description

Cross-modal pair. The 'before' is archive optical and the 'after' is the radar that can actually be tasked through cloud and at night.

Resolution. Ground resolution is 0.300-0.509 m and varies per tile across five CRS, so it is carried per sample in resolution_m rather than claimed on the bands. The post-event SAR is requantised 8-bit amplitude: no scale factor recovers backscatter from it.

Splits. The release's own *_setlevel.txt lists: 2230 train, 660 test (its holdout). Three documented defects in them are corrected: one tile listed in both, one listed in neither, and one leaked into val. validation (647) is one whole event, la_palma-volcano, carved out of the 2230 train tiles, and split_original keeps the release's own value. The competition's unlabelled val/ split is the held-out half below.

Held-out half

bright-dfc2025_test.zip holds the 348 samples whose targets the publisher withheld. Same inputs, no target slots: it is there to be predicted on and submitted, and it belongs in no training mixture.

import os
from huggingface_hub import snapshot_download
from taco.ml import Dataset

root = snapshot_download("isp-uv-es/bright-dfc2025-taco", repo_type="dataset",
                         allow_patterns=["bright-dfc2025_test.zip", "bright-dfc2025_test.zip/*", "bright-dfc2025_test.zip/.tacocat/*"])
ds = Dataset(os.path.join(root, "bright-dfc2025_test.zip"))

Getting started

git clone --recursive https://github.com/OscarPellicer/taco
pip install -e "taco/python[ml]"     # builds the reader: C++23, CMake, Ninja, pkg-config, libcurl >= 7.83, OpenSSL >= 3

Read it straight from the Hub:

import os
from huggingface_hub import hf_hub_download, snapshot_download
from taco.ml import Dataset, plot_sample

path = hf_hub_download("isp-uv-es/bright-dfc2025-taco", "bright-dfc2025.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

ds = Dataset("bright-dfc2025.zip")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["pre_optical"].array.shape

Metadata without decoding anything:

import taco
taco.read("bright-dfc2025.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	pre_optical	raster	optical	3 band(s), render
input	post_sar	raster	sar	1 band(s), requantised
target	damage	mask	label_raster	4 classes

Licence

CC-BY-NC-SA-4.0

Terms of use:

• Non-commercial use only (CC-BY-NC-SA-4.0).
• Adaptations must be shared under CC-BY-NC-SA-4.0.
• Credit the original authors and the data sources listed below.

Required credits:

• BRIGHT (Chen et al.), IEEE GRSS Data Fusion Contest 2025
• Imagery: Maxar (now Vantor) Open Data Program, licensed under CC BY-NC 4.0
• SAR imagery: Capella Space Open Data and Umbra Open Data, licensed under CC BY 4.0
• La Palma optical imagery: © Instituto Geográfico Nacional de España, CC BY 4.0
• Hawaii optical imagery: NOAA Office for Coastal Management

Providers: RIKEN AIP / University of Tokyo (Chen et al.)

Acknowledgements

TACO was designed by César Aybar and is specified at https://asterisk.coop/taco/spec/.

Built by Oscar Pellicer within the ELLIOT project at the Image and Signal Processing Group (ISP), Universitat de València.