StanfordExtra

12k keypoint and segmentation labelled instances of dogs in-the-wild.

Usage

To understand how the dataset can be used, please read demo.ipynb.

Installation

• All annotations, segmentations and metadata are sourced in a single .json file for ease of download. However, you will also need to download the Stanford Dogs Dataset dataset to access the raw images.

• For segmentation decoding, install pycocotools

python -m pip install "git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI"

• The demo.ipynb code is trivial to adapt to work with the full StanfordExtra dataset, by editing the line

img_dir = "sample_imgs" # Edit this to the location of the extracted tar file (e.g. /.../Images).

Download & Versioning

The latest version of StanfordExtra is available for download within this repo. To view all released versions, view the archive.

• Version 1.0 [23/08/20] - Initial release for ECCV 2020, 12k instances
• Version 0.1 [13/08/20] - Beta release, 11k instances

Comments

You may also find the other datasets useful for your animal work:

• Animal Pose Dataset
• RGBD-Dog Dataset
• BADJA

Acknowledgements

If you make sure use of this annotation dataset, please cite the following paper:

@inproceedings{biggs2020wldo,
  title={{W}ho left the dogs out: {3D} animal reconstruction with expectation maximization in the loop},
  author={Biggs, Benjamin and Boyne, Oliver and Charles, James and Fitzgibbon, Andrew and Cipolla, Roberto},
  booktitle={ECCV},
  year={2020}
}

and the Stanford Dog Dataset from which this is derived:

@inproceedings{KhoslaYaoJayadevaprakashFeiFei_FGVC2011,
author = "Aditya Khosla and Nityananda Jayadevaprakash and Bangpeng Yao and Li Fei-Fei",
title = "Novel Dataset for Fine-Grained Image Categorization",
booktitle = "First Workshop on Fine-Grained Visual Categorization, IEEE Conference on Computer Vision and Pattern Recognition",
year = "2011",
month = "June",
address = "Colorado Springs, CO",
}

Licensing

Non-commercial use only. Please contact us if you wish to use this dataset for commercial purposes. Data is provided `As-is', we take no liability for any errors.