Temporal Action Detection with Dynamic Weights Based on Curriculum Learning
Chen YZ(陈云泽)1,2; He jiang1,2; Junrui Xiao1,2; Ding Li1,2; Qingyi Gu1
Source PublicationNeurocomputing
2023
Pages106-116
Abstract

To enable temporal action localization, the computer needs to recognize the locations and classes of action instances in a video. The main challenge to temporal action detection is that the videos are often long and untrimmed, consisting of varying action content. Existing temporal action detection frameworks exhibit a gap between the training and testing phases, which is detrimental to model performance. Specifically, all positive samples are trained identically in the training phase. By contrast, in the testing phase, the positive samples with the best classification and localization scores are selected, while all others are suppressed. To mitigate this issue, we build an auxiliary branch to unify the training and testing procedures. In the construction of the auxiliary branch, we design a dynamic weighting strategy based on curriculum learning, where the weights of training samples are a combination of their classification and localization scores. Motivated by the speculation of curriculum learning, we emphasize the importance of classification and localization scores in different training stages. The classification score accounts for a higher proportion of the combined score in the early stages of the training process. As the epoch increases, the localization score gradually increases in proportion as well. The experimental results demonstrate that our methodology of curriculum-based learning enhances the performance of current action localization techniques. On THUMOS14, our technique outperforms the existing state of-the-art technique (57.6% vs 55.5%). And the performance on ActivityNet v1.3 (mAP@Avg) reaches 35.4%.

Indexed BySCI
Language英语
WOS IDWOS:000990128300001
Sub direction classification图像视频处理与分析
planning direction of the national heavy laboratory视觉信息处理
Paper associated data
Citation statistics
Document Type期刊论文
Identifierhttp://ir.ia.ac.cn/handle/173211/52385
Collection中科院工业视觉智能装备工程实验室_精密感知与控制
Affiliation1.Institute of Automation, Chinese Academy of Sciences, East Zhongguancun Road, Haidian District, Beijing, China
2.School of Artificial Intelligence, University of Chinese Academy of Sciences, Jingjia Road, Huairou District, Beijing, China
First Author AffilicationInstitute of Automation, Chinese Academy of Sciences
Recommended Citation
GB/T 7714
Chen YZ,He jiang,Junrui Xiao,et al. Temporal Action Detection with Dynamic Weights Based on Curriculum Learning[J]. Neurocomputing,2023:106-116.
APA Chen YZ,He jiang,Junrui Xiao,Ding Li,&Qingyi Gu.(2023).Temporal Action Detection with Dynamic Weights Based on Curriculum Learning.Neurocomputing,106-116.
MLA Chen YZ,et al."Temporal Action Detection with Dynamic Weights Based on Curriculum Learning".Neurocomputing (2023):106-116.
Files in This Item: Download All
File Name/Size DocType Version Access License
Temporal action dete(1252KB)期刊论文作者接受稿开放获取CC BY-NC-SAView Download
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[Chen YZ(陈云泽)]'s Articles
[He jiang]'s Articles
[Junrui Xiao]'s Articles
Baidu academic
Similar articles in Baidu academic
[Chen YZ(陈云泽)]'s Articles
[He jiang]'s Articles
[Junrui Xiao]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[Chen YZ(陈云泽)]'s Articles
[He jiang]'s Articles
[Junrui Xiao]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: Temporal action detection with dynamic weights based on curriculum.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.