isp-uv-es/quakeset-taco

QuakeSet

This is a repackaging, not a new dataset. It is QuakeSet by Politecnico di Torino (Rege Cambrin and Garza), ESA Copernicus Sentinel-1, 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-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:

@inproceedings{rege_cambrin2024quakeset,
  title     = {QuakeSet: A Dataset and Low-Resource Models to Monitor Earthquakes through Sentinel-1},
  author    = {Rege Cambrin, Daniele and Garza, Paolo},
  booktitle = {Proceedings of the International ISCRAM Conference (ISCRAM 2024)},
  year      = {2024},
  doi       = {10.59297/n89yc374}
}

About the data

3327 bitemporal Sentinel-1 GRD pairs of 512x512 at 10 m over 155 earthquakes, for earthquake detection and magnitude estimation: 1906 positive (pre-event, post-event) pairs carrying the event magnitude (M 4.45-6.87) and 1421 hard-negative (before, pre) pairs over the same ground.

3,327 samples · splits: test 511 · train 2,266 · validation 550 · tasks: change-detection, regression

Packaged as TACO v3.

Full description

Radiometry. Backscatter is linear power, established from the pixel distribution. Sentinel-1 gaps are kept as NaN rather than zeroed: in linear power a 0 is a perfect absorber, not a neutral filler.

Georeferencing. Each patch is placed by reconstructing its position in its event's mosaic, row-major, verified against the seams between adjacent patches. The release ships only bare coordinate axes.

Caveat. In 1,040 of the 1,421 negative pairs (73.2%) the two frames are bit-identical, so most of the negative class is separable by an equality test rather than by radar reasoning. frames_identical flags every one of them.

Splits. Official and event-disjoint: 2266 train, 550 val, 511 test pairs.

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/quakeset-taco", "quakeset.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

ds = Dataset("quakeset.zip")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["t1"].array.shape

Metadata without decoding anything:

import taco
taco.read("quakeset.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	t1	raster	sar	2 band(s), unit 1, physical
input	t2	raster	sar	2 band(s), unit 1, physical
target	affected	class_index		2 classes
target	magnitude	scalar		unit Mw

Licence

CC-BY-NC-4.0

Terms of use:

• Non-commercial use only (CC-BY-NC-4.0).

Required credits:

• QuakeSet: Cambrin and Garza (Politecnico di Torino)
• Contains modified Copernicus Sentinel-1 data

Providers: Politecnico di Torino (Rege Cambrin and Garza), ESA Copernicus Sentinel-1

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.