Search

Compared with face recognition, the performance of person re-identification (re-ID) is still far from practical application. Among various interferences, there are two factors seriously limiting the performance improvement, i.e., the feature discriminability determined by “external network effectiveness”, and the image quality determined by “internal background clutters”. Target at the “external network effectiveness” problem, feature pyramids are effective to learn discriminative features because they can learn both detailed features from high-resolution shallow layers and semantical features from low-resolution deep layers, however, it can only achieve slight improvement on re-ID tasks because of the error back propagation problem. To handle the problem and utilize the effectiveness of feature pyramids, we propose a strategy called Feature Pyramid Optimization (FPO). Instead of concatenating features directly, the selected layers are optimized independently in a top–bottom order. Target at the “internal background clutters” problem, background suppression is generally considered for removing the environmental interference and improving the image quality. Several mask-based methods are used attempting to totally remove background clutters but achieve limited promotion because of the mask sharpening effect. We propose a novel strategy, i.e., Gradual Background Suppression (GBS) to reduce the background clutters and keep the smoothness of images simultaneously. Extensive experiments have been conducted and the results demonstrate the effectiveness of both FPO and GBS.

Person re-identification (re-ID) is a technique aiming to recognize person cross different cameras. Although some supervised methods have achieved favorable performance, they are far from practical application owing to the lack of labeled data. Thus, unsupervised person re-ID methods are in urgent need. Generally, the commonly used approach in existing unsupervised methods is to first utilize the source image dataset for generating a model in supervised manner, and then transfer the source image domain to the target image domain. However, images may lose their identity information after translation, and the distributions between different domains are far away. To solve these problems, we propose an image domain-to-domain translation method by keeping pedestrian's identity information and pulling closer the domains' distributions for unsupervised person re-ID tasks. Our work exploits the CycleGAN to transfer the existing labeled image domain to the unlabeled image domain. Specially, a Self-labeled Triplet Net is proposed to maintain the pedestrian identity information, and maximum mean discrepancy is introduced to pull the domain distribution closer. Extensive experiments have been conducted and the results demonstrate that the proposed method performs superiorly than the state-ofthe- art unsupervised methods on DukeMTMC-reID and Market- 1501.