Journal / Conference
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR, 2020)
[PDF link: link]
[Code link: link]
Keywords
Person Search, Siamese RPN, Relation Block
Abstract
Person detection networks have been widely used in person search. These detectors discriminate persons from the background and generate proposals of all the persons from a gallery of scene images for each query. However, such a large number of proposals have a negative influence on the following identity matching process because many distractors are involved. In this paper, we propose a new detection network for person search, named Instance Guided Proposal Network (IGPN), which can learn the similarity between query persons and proposals. Thus, we can decrease proposals according to the similarity scores. To incorporate information of the query into the detection network, we introduce the Siamese region proposal network to Faster-RCNN and we propose improved cross-correlation layers to alleviate the imbalance of parameters distribution. Furthermore, we design a local relation block and a global relation branch to leverage the proposal-proposal relations and query-scene relations, respectively. Extensive experiments show that our method improves the person search performance through decreasing proposals and achieves competitive performance on two large person search benchmark datasets, CUHK-SYSU and PRW.
Method/Framework
The proposed IGPN mainly consists of an improved Siamese region proposal network (Siamese-RPN) and a local relation block which leverages appearance information of the query and relationships between pairs of proposals in the same scene, respectively. Moreover, it can exploit the global relationship between the query and scenes through the global relation branch. It takes a pair of a query person patch and a scene image as input and outputs bounding boxes along with similarity scores. When taken as a person detection network, IGPN works with a separately trained person Re-id network. Given a query person and a set of gallery scene images, we first obtain many proposals through IGPN. Then we remove the proposals with low similarity scores. Only the remaining ones are fed into the Re-id network.
Experiments
We perform several analytic experiments on CUHK-SYSU and PRW to explore the contribution of each component in our proposed IGPN.
Highlight
- We We propose a new person detection network, named Instance Guided Proposal Network (IGPN), which integrates the query person information into the detection network to learn the similarity between person proposals and the target person.
- We propose ICCL to alleviate the imbalance of parameters distribution in the vanilla Siamese-RPN without loss of performance.
- We design a local relation block and a global relation branch to leverage the proposal-proposal and query-scene relations, respectively.
Citation
@InProceedings{IGPN_2020_CVPR,
author = {Dong, Wenkai and Zhang, Zhaoxiang and Song, Chunfeng and Tan, Tieniu },
title = {Instance Guided Proposal Network for Person Search},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
year = {2020} }
