From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation
文献类型:期刊论文
作者 | Zhang, Zhaoxiang3,4,5,6; Luo, Chuanchen5,6; Wu, Haiping1; Chen, Yuntao1; Wang, Naiyan2; Song, Chunfeng5,6 |
刊名 | INTERNATIONAL JOURNAL OF COMPUTER VISION |
出版日期 | 2022-03-01 |
卷号 | 130期号:3页码:800-819 |
ISSN号 | 0920-5691 |
关键词 | Feature aggregation Deep learning Intra-class variance Person re-identification Video object detection |
DOI | 10.1007/s11263-021-01569-2 |
通讯作者 | Zhang, Zhaoxiang(zhaoxiang.zhang@ia.ac.cn) |
英文摘要 | The recording process of observation is influenced by multiple factors, such as viewpoint, illumination, and state of the object-of-interest etc.Thus, the image observation of the same object may vary a lot under different conditions. This leads to severe intra-class variance which greatly challenges the discrimination ability of the vision model. However, the current prevailing softmax loss for visual recognition only pursues perfect inter-class separation in the feature space. Without considering the intra-class compactness, the learned model easily collapses when it encounters the instances that deviate a lot from their class centroid. To resist the intra-class variance, we start by organizing the input instances as a graph. From this viewpoint, we find that the normalized cut on the graph is a favorable surrogate metric of the intra-class variance within the training batch. Inspired by the equivalence between the normalized cut and random walk, we propose a feature aggregation scheme using transition probabilities as guidance. By imposing supervision on the aggregated features, we can constrain the transition probabilities to form a graph partition consistent with the given labels. Thus, the normalized cut as well as intra-class variance can be well suppressed. To validate the effectiveness of this idea, we instantiate it in spatial, temporal, and spatial-temporal scenarios. Experimental results on corresponding benchmarks demonstrate that the proposed feature aggregation leads to significant improvement in performance. Our method is on par with, or even better than current state-of-the-arts in both tasks. |
WOS关键词 | PERSON REIDENTIFICATION ; ATTENTION ; ACCURATE ; NETWORK |
资助项目 | Major Project for New Generation of AI[2018AAA0100400] ; National Natural Science Foundation of China[61836014] ; National Natural Science Foundation of China[U21B2042] ; National Natural Science Foundation of China[61773375] ; National Natural Science Foundation of China[62006231] ; National Natural Science Foundation of China[62072457] ; National Youth Talent Support Program |
WOS研究方向 | Computer Science |
语种 | 英语 |
出版者 | SPRINGER |
WOS记录号 | WOS:000749450600003 |
资助机构 | Major Project for New Generation of AI ; National Natural Science Foundation of China ; National Youth Talent Support Program |
源URL | [http://ir.ia.ac.cn/handle/173211/47924] |
专题 | 自动化研究所_智能感知与计算研究中心 |
通讯作者 | Zhang, Zhaoxiang |
作者单位 | 1.Tusimple, San Diego, CA USA 2.TuSimple, Beijing, Peoples R China 3.Chinese Acad Sci HKISICAS, Hong Kong Inst Sci & Innovat, Ctr Artificial Intelligence & Robot, Hong Kong, Peoples R China 4.Chinese Acad Sci, Ctr Excellence Brain Sci & Intelligence Technol C, Beijing 100190, Peoples R China 5.Univ Chinese Acad Sci UCAS, Beijing 100190, Peoples R China 6.Chinese Acad Sci CASIA, Ctr Res Intelligent Percept & Comp CRIPAC, Natl Lab Pattern Recognit NLPR, Inst Automat, Beijing 100190, Peoples R China |
推荐引用方式 GB/T 7714 | Zhang, Zhaoxiang,Luo, Chuanchen,Wu, Haiping,et al. From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation[J]. INTERNATIONAL JOURNAL OF COMPUTER VISION,2022,130(3):800-819. |
APA | Zhang, Zhaoxiang,Luo, Chuanchen,Wu, Haiping,Chen, Yuntao,Wang, Naiyan,&Song, Chunfeng.(2022).From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation.INTERNATIONAL JOURNAL OF COMPUTER VISION,130(3),800-819. |
MLA | Zhang, Zhaoxiang,et al."From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation".INTERNATIONAL JOURNAL OF COMPUTER VISION 130.3(2022):800-819. |
入库方式: OAI收割
来源:自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。