Sequential Representation Learning via Static-Dynamic Conditional Disentanglement

TL;DR

This paper introduces a novel self-supervised method for disentangling static and dynamic factors in sequential data like videos, explicitly modeling their causal relationship and improving representation learning with normalizing flows.

Contribution

It proposes a new model that relaxes independence assumptions, incorporates causal relationships, and introduces a theoretically grounded disentanglement constraint for better sequential data representation.

Findings

Outperforms state-of-the-art methods in scene dynamics scenarios

Provides a formal definition and identifiability conditions for factors

Enhances model expressivity with Normalizing Flows

Abstract

This paper explores self-supervised disentangled representation learning within sequential data, focusing on separating time-independent and time-varying factors in videos. We propose a new model that breaks the usual independence assumption between those factors by explicitly accounting for the causal relationship between the static/dynamic variables and that improves the model expressivity through additional Normalizing Flows. A formal definition of the factors is proposed. This formalism leads to the derivation of sufficient conditions for the ground truth factors to be identifiable, and to the introduction of a novel theoretically grounded disentanglement constraint that can be directly and efficiently incorporated into our new framework. The experiments show that the proposed approach outperforms previous complex state-of-the-art techniques in scenarios where the dynamics of a scene are influenced by its content.