isp-uv-es/chatearthnet-taco

ChatEarthNet

This is a repackaging, not a new dataset. It is ChatEarthNet by Yuan, Xiong, Mou and Zhu (TUM), ESA Copernicus, 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-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{yuan2025chatearthnet,
  title   = {ChatEarthNet: a global-scale image--text dataset empowering vision--language geo-foundation models},
  author  = {Yuan, Zhenghang and Xiong, Zhitong and Mou, Lichao and Zhu, Xiao Xiang},
  journal = {Earth System Science Data},
  volume  = {17},
  pages   = {1245--1263},
  year    = {2025},
  doi     = {10.5194/essd-17-1245-2025}
}

About the data

163488 Sentinel-2 patches at 256x256 with a land-cover mask on an 11-class ESA WorldCover legend, 159348 co-registered Sentinel-1 VV/VH pairs, and GPT-3.5 and GPT-4V captions describing the land cover.

163,488 samples · splits: test 47,816 · train 99,760 · validation 15,912 · tasks: semantic-segmentation, image-captioning

Packaged as TACO v3.

Full description

Radiometry. The imagery is a per-image contrast-stretched 8-bit PNG render, not reflectance. Every patch is stretched independently, so channel values are not comparable between patches and no affine map back to a physical unit exists. All pixel slots are render and carry no approx_scale. No georeferencing is published, so none is written.

Legend. Not shipped by the release. Exactly 12 colours occur over all 163488 masks (10.71e9 pixels), mapped to classes by exclusive-mention and rare-class prevalence tests against the captions, which were written from the label raster; separations range 6x-400x. WorldCover's snow_and_ice does not occur.

Splits. The two caption sets ship two official splits that disagree on 5308 of the 9872 patches they share, so both are carried.

• split is the GPT-3.5 division (159348 patches) and split_4v the GPT-4V one (9872). Merging them would move ~54% of the 4V test set into training.
• The release partitions the uncaptioned patches nowhere. They are written train, held out by neither official split, with the release's own word kept in split_original.

Radar. Sentinel-1 is a variable leaf, present on exactly the 159348 captioned patches. The remaining 4140 have imagery and a mask but no radar and no caption. S1 is never zero-filled.

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:

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

path = snapshot_download("isp-uv-es/chatearthnet-taco", repo_type="dataset")
ds = Dataset(path)                  # the repository is the container: parts + .tacocat
plot_sample(ds[0])

or from a local copy:

ds = Dataset("chatearthnet")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["s2_rgb"].array.shape

Metadata without decoding anything:

import taco
taco.read("chatearthnet/.tacocat")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	s2_rgb	raster	optical	3 band(s), render
input	s2_b567	raster	optical	3 band(s), render
input	s2_b81112	raster	optical	3 band(s), render
input	s1	raster_series	sar	2 band(s), render
input	caption_35	text	text
input	caption_4v	text	text
target	land_cover	mask	label_raster	11 classes
target	caption	text	text

Licence

CC-BY-4.0

Providers: Yuan, Xiong, Mou and Zhu (TUM), ESA Copernicus

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.