Disentangling Video with Independent Prediction

TL;DR

This paper introduces an unsupervised variational model that disentangles videos into independent, interpretable factors, enabling future prediction of each factor from its past without interference from others.

Contribution

The paper presents a novel unsupervised variational approach for video disentanglement that produces interpretable factors as objects in scenes.

Findings

Factors are often interpretable as scene objects.

The model successfully predicts future states of individual factors.

Disentanglement improves understanding of scene dynamics.

Abstract

We propose an unsupervised variational model for disentangling video into independent factors, i.e. each factor's future can be predicted from its past without considering the others. We show that our approach often learns factors which are interpretable as objects in a scene.