SRG: Snippet Relatedness-based Temporal Action Proposal Generator

TL;DR

This paper introduces SRG, a novel snippet relatedness-based method for temporal action proposal generation that effectively captures long-range dependencies, leading to improved proposals and action detection performance.

Contribution

The paper proposes a new snippet score-based approach with pyramid non-local operations to learn snippet relatedness for better temporal proposals.

Findings

SRG outperforms state-of-the-art methods on THUMOS-14 and ActivityNet-1.3.

SRG significantly improves temporal action detection accuracy.

The method effectively captures long-range snippet dependencies.

Abstract

Recent temporal action proposal generation approaches have suggested integrating segment- and snippet score-based methodologies to produce proposals with high recall and accurate boundaries. In this paper, different from such a hybrid strategy, we focus on the potential of the snippet score-based approach. Specifically, we propose a new snippet score-based method, named Snippet Relatedness-based Generator (SRG), with a novel concept of "snippet relatedness". Snippet relatedness represents which snippets are related to a specific action instance. To effectively learn this snippet relatedness, we present "pyramid non-local operations" for locally and globally capturing long-range dependencies among snippets. By employing these components, SRG first produces a 2D relatedness score map that enables the generation of various temporal intervals reliably covering most action instances with high overlap. Then, SRG evaluates the action confidence scores of these temporal intervals and refines their boundaries to obtain temporal action proposals. On THUMOS-14 and ActivityNet-1.3 datasets, SRG outperforms state-of-the-art methods for temporal action proposal generation. Furthermore, compared to competing proposal generators, SRG leads to significant improvements in temporal action detection.