开源软件名称（OpenSource Name）：

开源软件地址(OpenSource Url)：

开源编程语言(OpenSource Language)：

开源软件介绍(OpenSource Introduction)：

To be Updated for AAAI'18

SAP: Self-Adaptive Proposal Model for Temporal Action Detection based on Reinforcement Learning

By Jingjia Huang, Nannan Li, Tao Zhang and Ge Li

The paper can be found here.

Introduction

Self-Adaptive Proposal (SAP) is a DQN based model for temporal action localization in untrimmed long videos. The temporal action detection process for SAP is naturally one of observation and refinement: observe the current window and refine the span of attended window to cover true action regions. SAP can learn to find actions through continuously adjusting the temporal bounds in a self-adaptive way. Experiment results on THUMOS’14 validate the effectiveness of SAP, which can achieve competitive performance with current action detection algorithms via much fewer proposals.

fig.1 SAP architecture

fig.2 Illustration of DQN actions.Each yellow window with dashed lines represents the next window after taking the corresponding action.

fig.3 Example of how SAP works

This code has been tested on Ubuntu 16.04 with NVIDIA Tesla K80. The CUDA version is 8.0.61

License

SAP is released under the MIT License.

Citing

If you find SAP useful, please consider citing:

@paper{AAAI1816109,
	author = {Jingjia Huang and Nannan Li and Tao Zhang and Ge Li and Tiejun Huang and Wen Gao},
	title = {SAP: Self-Adaptive Proposal Model for Temporal Action Detection Based on Reinforcement Learning},
	conference = {AAAI Conference on Artificial Intelligence},
	year = {2018},
	keywords = {Computer vision; Action detection; Reinforcement learning},
	url = {https://www.aaai.org/ocs/index.php/AAAI/AAAI18/paper/view/16109}
}

If you like this project, give us a ⭐ in the github banner