isp-uv-es/everwatch-taco

EverWatch (Everglades wading bird detection)

This is a repackaging, not a new dataset. It is EverWatch (Everglades wading bird detection) by Weecology / University of Florida (Everglades Wading Bird Project), 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: CC0-1.0

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

Citation

Please cite the original work:

@dataset{garner2024everwatch,
  title     = {EverWatch benchmark: training and evaluation data for detection and species classification of Everglades wading birds from airborne imagery},
  author    = {Garner, Lindsey and Weinstein, Ben and Rickershauser, Michael and Baldino, Melissa and Coates, Holly and Commins, Mary and Faber, Tracey and Gula, Jonah and Van Ert, Shayly and Frederick, Peter and White, Ethan and Ernest, S. K. Morgan},
  publisher = {Zenodo},
  version   = {v0.1.0},
  year      = {2024},
  doi       = {10.5281/zenodo.11165946}
}

About the data

6127 airborne RGB orthophotos of roughly 1500x1500 px over Everglades wading-bird colonies, with 54602 bounding boxes over 8 species.

6,127 samples · splits: test 196 · train 4,994 · validation 937 · tasks: object-detection, counting

Packaged as TACO v3.

Full description

Boxes. Coordinates are in pixels: the imagery is distributed without georeferencing, so no CRS is declared. Per-class counts are the scientific product here, so counting is declared as a task alongside detection.

Encoding. Converted from the original PNGs and stored byte-verbatim: PNG is lossless, and re-encoding costs 35% more bytes.

Splits. The release's own train/test split is preserved. It ships no validation part, so one is carved out of train on whole colony visits (site plus flight date, 3,118 of them across the images, none straddling the release's own line), never out of test.

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

or from a local copy:

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

Metadata without decoding anything:

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

Samples

What a sample contains

role	slot	holds	modality	detail
input	image	raster	optical	3 band(s), render
target	boxes	bbox_2d
target	species	class_sequence		8 classes

Licence

CC0-1.0

Providers: Weecology / University of Florida (Everglades Wading Bird Project)

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.