stockeh/dog-pose-cv

Dataset Card for DogPoseCV

This dataset contains 20,578 images of dogs in various poses, labeled as standing, sitting, lying down, or undefined. It is intended for computer vision tasks to identify a dog's behavior from images.

Dataset Details

• Curated by: Jason Stock and Tom Cavey, Computer Science, Colorado State University
• Paper: arxiv.org/abs/2101.02380 (BibTeX)
• Repository: github.com/stockeh/canine-embedded-ml

The dataset is intended to be used to train computer vision models to identify a dog's pose/behavior (standing, sitting, lying down) from images. This can enable applications to automatically detect and respond to a dog's actions. The variety of dog breeds enables robust generalization for real-time inference of dog actions.

Dataset Structure

The dataset contains 20,578 RGB images of 120 dog breeds. Images are labeled as one of four classes:

• standing (4143 images)
• sitting (3038 images)
• lying down (7090 images)
• undefined (6307 images)

Images have varying resolutions, with 50% between 361x333 and 500x453 pixels.

Data Collection and Processing

This dataset is an adaption of from the Stanford Dog Dataset, relabeling dog breeds to their associated position. We manually labeled each image as standing, sitting, lying down, or undefined if the pose was indistinguishable, e.g., between two positions.

Bias, Risks, and Limitations

The dataset has a class imbalance, with nearly 2x as many "lying down" images compared to "sitting". Indistinguishable poses were labeled as "undefined", with most being close-up portraits. This may limit the ability to handle such images.

Recommendations: When using this dataset, be aware of the class imbalance and consider oversampling or augmentation techniques..

Citation

@article{stock2021s,
  title={Who's a Good Boy? Reinforcing Canine Behavior in Real-Time using Machine Learning},
  author={Stock, Jason and Cavey, Tom},
  journal={arXiv preprint arXiv:2101.02380},
  year={2021}
}