srvp-pretrained-model-mirror

SRVP Pretrained Models

This repository contains pretrained models for Stochastic Latent Residual Video Prediction (SRVP), originally developed by Jean-Yves Franceschi, Edouard Delasalles, Mickael Chen, Sylvain Lamprier, and Patrick Gallinari. This is a mirror of the official pretrained models for easier access and preservation.

Original Work

• Paper: Stochastic Latent Residual Video Prediction (ICML 2020)
• Original Repository: edouardelasalles/srvp
• Authors: Jean-Yves Franceschi, Edouard Delasalles, Mickael Chen, Sylvain Lamprier, Patrick Gallinari

Model Description

The model generates future video frames by learning a stochastic latent dynamics model that captures both deterministic motion and inherent uncertainty in future predictions.

Available Models

This repository contains pretrained models for all datasets mentioned in the original paper:

1. Stochastic Moving MNIST
2. Deterministic Moving MNIST
3. KTH Actions
4. Human3.6M
5. BAIR Robot Pushing

Usage

Please refer to the original repository for detailed usage instructions and code implementation.

Limitations

• Performance depends on the similarity between test data and training data
• May struggle with highly complex scenes or long-term predictions
• Computational requirements can be significant for high-resolution videos

Citation

If you use these models, please cite the original paper:

@inproceedings{franceschi2020stochastic,
  title={Stochastic latent residual video prediction},
  author={Franceschi, Jean-Yves and Delasalles, Edouard and Chen, Micka{\"e}l and Lamprier, Sylvain and Gallinari, Patrick},
  booktitle={International Conference on Machine Learning},
  pages={3233--3246},
  year={2020},
  organization={PMLR}
}

Original Work

• Paper: Stochastic Latent Residual Video Prediction (ICML 2020)
• Original Repository: edouardelasalles/srvp
• Authors: Jean-Yves Franceschi, Edouard Delasalles, Mickael Chen, Sylvain Lamprier, Patrick Gallinari

License

This model mirror is released under Apache 2.0 license, the same as the original repository.

Acknowledgements

Thanks to the original authors for making their models publicly available. This is an unofficial mirror created for preservation and easier access through Hugging Face.