CryptoThaler/DeepLabCutModelZoo-SuperAnimal-Quadruped

SuperAnimal-Quadruped research demo

This is an unofficial deployment of the DeepLabCut SuperAnimal-Quadruped model for research demonstrations. Upload a short quadruped video to generate a labeled video with predicted anatomical keypoints.

Model and author credit

The SuperAnimal method, Quadruped-80K dataset, and SuperAnimal-Quadruped model were developed by Shaokai Ye, Anastasiia Filippova, Jessy Lauer, Steffen Schneider, Maxime Vidal, Tian Qiu, Alexander Mathis, and Mackenzie Weygandt Mathis at the Mathis Laboratory of Adaptive Intelligence / DeepLabCut team.

• —Official model and weights: mwmathis/DeepLabCutModelZoo-SuperAnimal-Quadruped
• —DeepLabCut software: DeepLabCut/DeepLabCut
• —This Space uses the hrnet_w32 pose model with the fasterrcnn_resnet50_fpn_v2 detector.

Paper and citation

Ye, S., Filippova, A., Lauer, J. et al. “SuperAnimal pretrained pose estimation models for behavioral analysis.” Nature Communications 15, 5165 (2024). https://doi.org/10.1038/s41467-024-48792-2

@article{ye2024superanimal,
  title={SuperAnimal pretrained pose estimation models for behavioral analysis},
  author={Ye, Shaokai and Filippova, Anastasiia and Lauer, Jessy and Schneider, Steffen and Vidal, Maxime and Qiu, Tian and Mathis, Alexander and Mathis, Mackenzie Weygandt},
  journal={Nature Communications},
  volume={15},
  pages={5165},
  year={2024},
  doi={10.1038/s41467-024-48792-2}
}

Use and limitations

The model is intended for academic and non-commercial research. Predictions should be reviewed before quantitative behavioral analysis, particularly for species, viewpoints, occlusions, and environments that differ from the model's training data. This Space is not affiliated with or endorsed by the model authors.

The first inference can take longer while DeepLabCut downloads and caches the official pose and detector checkpoints.