Activity Graph Transformer for Temporal Action Localization

TL;DR

The paper introduces Activity Graph Transformer, a novel end-to-end model that uses graph reasoning to improve temporal action localization in videos, especially for non-linear and overlapping actions.

Contribution

It proposes a graph-based transformer model for directly predicting action instances, addressing limitations of sequential processing in complex video scenarios.

Findings

Outperforms state-of-the-art on THUMOS14, Charades, and EPIC-Kitchens-100 datasets.

Effectively captures non-linear temporal dependencies and overlapping actions.

Demonstrates significant accuracy improvements over existing methods.

Abstract

We introduce Activity Graph Transformer, an end-to-end learnable model for temporal action localization, that receives a video as input and directly predicts a set of action instances that appear in the video. Detecting and localizing action instances in untrimmed videos requires reasoning over multiple action instances in a video. The dominant paradigms in the literature process videos temporally to either propose action regions or directly produce frame-level detections. However, sequential processing of videos is problematic when the action instances have non-sequential dependencies and/or non-linear temporal ordering, such as overlapping action instances or re-occurrence of action instances over the course of the video. In this work, we capture this non-linear temporal structure by reasoning over the videos as non-sequential entities in the form of graphs. We evaluate our model on challenging datasets: THUMOS14, Charades, and EPIC-Kitchens-100. Our results show that our proposed model outperforms the state-of-the-art by a considerable margin.