Dynamic information enhancement for video classification

作者：

Highlights：

• Explicitly equipping 2D models with the ability modeling bidirectional motion.

• Adaptively adjusting information, considering both temporal and spatial clues.

• Achieving a video recognition accuracy with 2D backbone comparable to SOTA methods.

摘要

Highlights•Explicitly equipping 2D models with the ability modeling bidirectional motion.•Adaptively adjusting information, considering both temporal and spatial clues.•Achieving a video recognition accuracy with 2D backbone comparable to SOTA methods.

论文关键词：Video classification,Spatiotemporal modelling,Explicitly encoding,Adaptive excitation

论文评审过程：Received 24 January 2021, Revised 28 May 2021, Accepted 7 June 2021, Available online 15 July 2021, Version of Record 14 August 2021.

论文官网地址：https://doi.org/10.1016/j.imavis.2021.104244