UnweaveNet: Unweaving Activity Stories

TL;DR

UnweaveNet is a neural model designed to parse egocentric videos into distinct activity threads, effectively detecting goal changes and activity resumption, with demonstrated success on the EPIC-KITCHENS dataset.

Contribution

The paper introduces UnweaveNet, a novel neural architecture with a thread bank and controller for unweaving complex activity sequences in videos, including a self-supervised pretraining approach.

Findings

Effective activity thread detection demonstrated on EPIC-KITCHENS.

Self-supervised pretraining improves model performance.

UnweaveNet accurately identifies goal changes and activity resumption.

Abstract

Our lives can be seen as a complex weaving of activities; we switch from one activity to another, to maximise our achievements or in reaction to demands placed upon us. Observing a video of unscripted daily activities, we parse the video into its constituent activity threads through a process we call unweaving. To accomplish this, we introduce a video representation explicitly capturing activity threads called a thread bank, along with a neural controller capable of detecting goal changes and resuming of past activities, together forming UnweaveNet. We train and evaluate UnweaveNet on sequences from the unscripted egocentric dataset EPIC-KITCHENS. We propose and showcase the efficacy of pretraining UnweaveNet in a self-supervised manner.