Temporal Action Localization : A Review

Temporal action localization(TAL) is a fundamental and extremely challenging problem in video understanding in computer vision. The growth of online media has led to the continuous generation of a vast amount of untrimmed videos, which has resulted in substantial research attention on TAL in recent years. This article presents a review of the main advances in the TAL field, covering both traditional methods and deep learning techniques. The article also reviews emerging research findings in areas such as weakly-supervised learning and multimodal data processing, analyzing and contrasting the advantages of these models. This paper presents commonly used datasets and evaluation metrics that are employed in this task. Finally, this article summarizes the current challenges in the field of TAL and suggests future directions for the development of TAL research.