Self-Regulated Learning for Egocentric Video Activity Anticipation

TL;DR

This paper introduces a Self-Regulated Learning framework for egocentric video activity anticipation that improves prediction accuracy by emphasizing novel information and correlating current frames with past observations, outperforming existing methods.

Contribution

The proposed SRL framework effectively regulates intermediate representations using contrastive loss and dynamic reweighing, enhancing activity anticipation in egocentric videos.

Findings

SRL outperforms state-of-the-art methods on multiple datasets.

The framework accurately identifies supporting action and object concepts.

Multi-task learning further improves activity representation quality.

Abstract

Future activity anticipation is a challenging problem in egocentric vision. As a standard future activity anticipation paradigm, recursive sequence prediction suffers from the accumulation of errors. To address this problem, we propose a simple and effective Self-Regulated Learning framework, which aims to regulate the intermediate representation consecutively to produce representation that (a) emphasizes the novel information in the frame of the current time-stamp in contrast to previously observed content, and (b) reflects its correlation with previously observed frames. The former is achieved by minimizing a contrastive loss, and the latter can be achieved by a dynamic reweighing mechanism to attend to informative frames in the observed content with a similarity comparison between feature of the current frame and observed frames. The learned final video representation can be further enhanced by multi-task learning which performs joint feature learning on the target activity labels and the automatically detected action and object class tokens. SRL sharply outperforms existing state-of-the-art in most cases on two egocentric video datasets and two third-person video datasets. Its effectiveness is also verified by the experimental fact that the action and object concepts that support the activity semantics can be accurately identified.