themattinthehatt / daart

Deep learning for animal action recognition toolbox

daart: deep learning for animal action recognition toolbox

A collection of tools for the discrete classification of animal behaviors using low-dimensional representations of videos (such as skeletons provided by tracking algorithms). Our approach combines strong supervision, weak supervision, and self-supervision to improve model performance. See the preprint here for more details.

This repo currently supports fitting the following types of base models on behavioral time series data:

• Dense MLP network with initial 1D convolutional layer
• RNNs - both LSTMs and GRUs
• Temporal Convolutional Networks (TCNs)

See the documentation to get started!

If you use daart in your analysis of behavioral data, please cite our preprint!

@inproceedings{whiteway2021semi,
  title={Semi-supervised sequence modeling for improved behavioral segmentation},
  author={Whiteway, Matthew R and Schaffer, Evan S and Wu, Anqi and Buchanan, E Kelly and Onder, Omer F and Mishra, Neeli and Paninski, Liam},
  journal={bioRxiv},
  year={2021},
  publisher={Cold Spring Harbor Laboratory}
}