Manipulation-Skill Assessment from Videos with Spatial Attention

Code for EPIC@ICCV2019 paper: Manipulation-skill assessment from videos with spatial attention network

Introduction

Quatifying and ranking the action quality performed in videos by decaying the attention to the redundent background parts and modeling temporal relationships on key parts.

We proposed a novel RNN-based spatial attention model that considers accumulated attention state from previous frames as well as high-level information about the progress of an undergoing task.

	>
	>

The approach is validated on four existing datasets of hand manipulation tasks, including Surgery, Drawing, Cooking and Using Chopsticks.

Reference

@inproceedings{li2019manipulation,
  title={Manipulation-skill assessment from videos with spatial attention network},
  author={Li, Zhenqiang and Huang, Yifei and Cai, Minjie and Sato, Yoichi},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops},
  pages={0--0},
  year={2019}
}