Unsupervised Video Representation Learning by Bidirectional Feature Prediction

TL;DR

This paper proposes a self-supervised video representation learning method that predicts both past and future features in a contrastive framework, improving action recognition performance by leveraging bidirectional temporal signals.

Contribution

It introduces a novel bidirectional feature prediction approach that combines past and future signals for enhanced self-supervised video learning.

Findings

Outperforms methods predicting only future or past features.

Enriches representations for action recognition tasks.

Utilizes contrastive learning with temporal hard negatives.

Abstract

This paper introduces a novel method for self-supervised video representation learning via feature prediction. In contrast to the previous methods that focus on future feature prediction, we argue that a supervisory signal arising from unobserved past frames is complementary to one that originates from the future frames. The rationale behind our method is to encourage the network to explore the temporal structure of videos by distinguishing between future and past given present observations. We train our model in a contrastive learning framework, where joint encoding of future and past provides us with a comprehensive set of temporal hard negatives via swapping. We empirically show that utilizing both signals enriches the learned representations for the downstream task of action recognition. It outperforms independent prediction of future and past.