Sequential Representation Learning via Static-Dynamic Conditional Disentanglement

Overview

This repository contains the implementation for the Sequential Representation Learning via Static-Dynamic Conditional Disentanglement paper which aims at self-supervised disentangled representation learning within sequential data, focusing on separating time-independent and time-varying factors in videos.

Training and testing

The provided files allow to train from scratch and test the proposed method on the dSprites dataset as presented in the paper. To run the training process:

./train_and_test.sh

Paper

@article{simon2024sequential,
  title={Sequential Representation Learning via Static-Dynamic Conditional Disentanglement},
  author={Simon, Mathieu Cyrille and Frossard, Pascal and De Vleeschouwer, Christophe},
  journal={arXiv preprint arXiv:2408.05599},
  year={2024}
}