Technical Report for ActivityNet Challenge 2022 -- Temporal Action Localization

TL;DR

This technical report presents a unified approach for temporal action localization in untrimmed videos, utilizing VideoSwinTransformer features, a Faster-TAD based network, and model ensembling to achieve high performance.

Contribution

It introduces a simplified, unified framework for temporal action detection that combines advanced feature extraction, a single-stage detection network, and model ensembling.

Findings

Achieved comparable results to multi-step methods.

Unified framework simplifies the TAD pipeline.

Ensembling improves detection accuracy.

Abstract

In the task of temporal action localization of ActivityNet-1.3 datasets, we propose to locate the temporal boundaries of each action and predict action class in untrimmed videos. We first apply VideoSwinTransformer as feature extractor to extract different features. Then we apply a unified network following Faster-TAD to simultaneously obtain proposals and semantic labels. Last, we ensemble the results of different temporal action detection models which complement each other. Faster-TAD simplifies the pipeline of TAD and gets remarkable performance, obtaining comparable results as those of multi-step approaches.