Visual Representation Learning with Stochastic Frame Prediction

TL;DR

This paper introduces a stochastic frame prediction framework with an auxiliary masked image modeling task to improve self-supervised learning of video representations, demonstrating effectiveness across various vision and robotics tasks.

Contribution

It proposes a novel stochastic video prediction approach combined with dense frame modeling for enhanced representation learning.

Findings

Effective in video label propagation

Improves pose tracking accuracy

Enhances robotic vision tasks

Abstract

Self-supervised learning of image representations by predicting future frames is a promising direction but still remains a challenge. This is because of the under-determined nature of frame prediction; multiple potential futures can arise from a single current frame. To tackle this challenge, in this paper, we revisit the idea of stochastic video generation that learns to capture uncertainty in frame prediction and explore its effectiveness for representation learning. Specifically, we design a framework that trains a stochastic frame prediction model to learn temporal information between frames. Moreover, to learn dense information within each frame, we introduce an auxiliary masked image modeling objective along with a shared decoder architecture. We find this architecture allows for combining both objectives in a synergistic and compute-efficient manner. We demonstrate the effectiveness of our framework on a variety of tasks from video label propagation and vision-based robot learning domains, such as video segmentation, pose tracking, vision-based robotic locomotion, and manipulation tasks. Code is available on the project webpage: https://sites.google.com/view/2024rsp.