End-to-end Temporal Action Detection with Transformer

Temporal action detection (TAD) aims to determine the semantic label and the boundaries of every action instance in an untrimmed video. It is a fundamental task in video understanding and significant progress has been made in TAD. Previous methods involve multiple stages or networks and hand-designed rules or operations, which fall short in efficiency and flexibility. Here, we construct an end-to-end framework for TAD upon Transformer, termed \textit{TadTR}, which simultaneously predicts all action instances as a set of labels and temporal locations in parallel. TadTR is able to adaptively extract temporal context information needed for making action predictions, by selectively attending to a number of snippets in a video. It greatly simplifies the pipeline of TAD and runs much faster than previous detectors. Our method achieves state-of-the-art performance on HACS Segments and THUMOS14 and competitive performance on ActivityNet-1.3. Our code will be made available at \url{https://github.com/xlliu7/TadTR}.