Disentangling Physical Dynamics from Unknown Factors for Unsupervised Video Prediction

TL;DR

This paper introduces PhyDNet, a novel deep learning architecture that disentangles physical PDE dynamics from unknown factors for improved unsupervised video prediction, demonstrating superior performance and robustness.

Contribution

The paper proposes PhyDNet, a two-branch deep architecture with a PDE-inspired recurrent cell, enabling explicit disentanglement of physical dynamics from unknown information in video prediction.

Findings

Outperforms state-of-the-art methods on four datasets.

Disentanglement and PDE-constrained prediction significantly improve results.

Effective in handling missing data and long-term forecasting.

Abstract

Leveraging physical knowledge described by partial differential equations (PDEs) is an appealing way to improve unsupervised video prediction methods. Since physics is too restrictive for describing the full visual content of generic videos, we introduce PhyDNet, a two-branch deep architecture, which explicitly disentangles PDE dynamics from unknown complementary information. A second contribution is to propose a new recurrent physical cell (PhyCell), inspired from data assimilation techniques, for performing PDE-constrained prediction in latent space. Extensive experiments conducted on four various datasets show the ability of PhyDNet to outperform state-of-the-art methods. Ablation studies also highlight the important gain brought out by both disentanglement and PDE-constrained prediction. Finally, we show that PhyDNet presents interesting features for dealing with missing data and long-term forecasting.