DigiDogs: A 3D Synthetic Dog Dataset

This dataset is a part of the following publication:

"DigiDogs: Single-View 3D Pose Estimation of Dogs using Synthetic Training Data"
Moira Shooter, Charles Malleson, Adrian Hilton
Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops, 2024, pp. 80-89
The work was presented at the CV4Smalls workshop

Paper - Code - Flash Talk

The data

We present a 3D dog video dataset which was generated using the popular game Grand Theft Auto V. We collected 118 videos, including 27,900 frames capturing a diverse range of scenarios and canine. In each video, the dogs performed either a single animation, like running, or a sequence of animations, such as walking followed by sitting. Alongside the RGB images, we generated depth maps, kinematic skeletal motion sequences, 2D/3D keypoint coordinates, segmentation maps and camera intrinsics and extrinsics.

The structure of the content is stated bellow:

  • 08_30_2022 to 09_26_2022 These folders contain subfolders which refer to the sequence ID e.g. 00001:

    • images: The images in format of .tiff (which contain the RGB, depth and segmentation map).
    • 2DCoordinates.log: The screen coordinates of the keypoints.
    • 3DCoordinates.log: The 3D camera coordinates of the keypoints in coordinate system of GTA-V (left).
    • CameraInfo.log: The camera information for each frame (trans, rot, fov, width, height, camera matrix).
    • MotionSequence.bvh: Motion capture sequence in .bvh format. This can be imported in any 3D software such as Blender, Autodesk Maya, etc.
  • split

    The dataset split can be found in the {trainpose,valpose,testpose}.pickle files, which represent the frames used. The files named {00001-00118}.json represent the sequences.

  • statistics.csv

    Here you can see which dog is used for each recorded sequence.

License Agreement

License file is in license.txt

  1. All original images and associated data provided may be used for non-commercial research purposes only.
  2. The source of the datasets must be acknowledged in all publications where they are used.
  3. The data may not be redistributed.

Download

All the data can be downloaded by clicking the following link: DigiDogs2024_full.tar.gz

If you use this dataset/code please cite:

@InProceedings{Shooter_2024_WACV,
author = {Shooter, Moira and Malleson, Charles and Hilton, Adrian},
title = {DigiDogs: Single-View 3D Pose Estimation of Dogs Using Synthetic Training Data},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops},
month = {January},
year = {2024},
pages = {80-89}
}

Questions:

For any questions about this dataset, please contact Moira Shooter.