Learning to Decompose and Disentangle Representations for Video Prediction

TL;DR

This paper introduces DDPAE, a novel framework that decomposes and disentangles video representations to improve future frame prediction, learning these features without supervision and handling complex multi-object interactions.

Contribution

The paper presents DDPAE, a new model combining probabilistic and deep learning methods to automatically decompose and disentangle video representations for prediction tasks.

Findings

Successfully decomposes moving digits into components and disentangles appearance and location.

Predicts future frames in complex multi-object videos without supervision.

Recovers underlying physical states from raw pixel data.

Abstract

Our goal is to predict future video frames given a sequence of input frames. Despite large amounts of video data, this remains a challenging task because of the high-dimensionality of video frames. We address this challenge by proposing the Decompositional Disentangled Predictive Auto-Encoder (DDPAE), a framework that combines structured probabilistic models and deep networks to automatically (i) decompose the high-dimensional video that we aim to predict into components, and (ii) disentangle each component to have low-dimensional temporal dynamics that are easier to predict. Crucially, with an appropriately specified generative model of video frames, our DDPAE is able to learn both the latent decomposition and disentanglement without explicit supervision. For the Moving MNIST dataset, we show that DDPAE is able to recover the underlying components (individual digits) and disentanglement (appearance and location) as we would intuitively do. We further demonstrate that DDPAE can be applied to the Bouncing Balls dataset involving complex interactions between multiple objects to predict the video frame directly from the pixels and recover physical states without explicit supervision.