Masked Autoencoders for Egocentric Video Understanding @ Ego4D Challenge 2022

TL;DR

This paper describes the application of masked autoencoders to egocentric video understanding tasks, achieving high rankings in the Ego4D Challenge 2022, with results on object state change classification and temporal localization.

Contribution

The paper introduces the use of masked autoencoders for egocentric video tasks and demonstrates their effectiveness through empirical results in a competitive challenge.

Findings

Ranked 2nd in Object State Change Classification

Ranked 2nd in PNR Temporal Localization

Code will be publicly available

Abstract

In this report, we present our approach and empirical results of applying masked autoencoders in two egocentric video understanding tasks, namely, Object State Change Classification and PNR Temporal Localization, of Ego4D Challenge 2022. As team TheSSVL, we ranked 2nd place in both tasks. Our code will be made available.