agentic-learning-ai-lab/Midway-Network

Midway Network: Learning Representations for Recognition and Motion from Latent Dynamics

Paper | Code| Website

These are trained models instantiating the Midway Network (ICLR 2026) architecture for self-supervised learning of visual representations for recognition and motion from videos.

Midway Network: Learning Representations for Recognition and Motion from Latent Dynamics
Christopher Hoang, Mengye Ren
International Conference on Learning Representations 2026
arXiv (arXiv 2510.05558)

The models are trained on BDD100K or WT-Venice.

Citation

If you find this repository useful in your research, please consider giving a like and a citation:

@inproceedings{hoang:2026:midway-network,
  title={Midway Network: Learning Representations for Recognition and Motion from Latent Dynamics}, 
    author={Chris Hoang and Mengye Ren},
  booktitle={International Conference on Learning Representations},  
  year={2026}
}