M-adapter: Multi-level image-to-video adaptation for video action recognition

Rongchang Li, Tianyang Xu, Xiaojun Wu, Linze Li, Xiao Yang, Zhongwei Shen, Josef Kittler

Abstract: Highlights•A parallel parameter-efficient tuning method with less cost than current methods.•A multi-level adapter that adapts pre-trained image models for action recognition.•Comprehensive comparisons between full-parameter and proposed tuning approach.•Competitive performance to SOTA results while requiring lower training costs.