Patch Based Discriminative Feature Learning for Unsupervised Person Re-identification


While discriminative local features have been shown effective in solving the person re-identification problem which aims to match pedestrians across non-overlapping camera views, they are currently limited to be trained on fully pairwise labelled data as far as we know. In this work, we overcome this problem by proposing a patch-based unsupervised learning framework in order to learn discriminative feature from patches instead of the whole images. The patch based learning method would be more able to leverage similar patches to learn a discriminative model. Specifically, we develop a PatchNet to select patches from the feature map and learn discriminative features for these patches. Furthermore, we propose a patch-based discriminative feature learning loss to provide effective guidance for the PatchNet in learning discriminative patch feature on the unlabeled dataset. Simultaneously, an image-level feature learning loss is designed to leverage all the patch features of the same image to serve as an image-level guidance for the PatchNet. Extensive experiments validate the superiority of our method for unsupervised person re-id.

International Conference on Computer Vision and Pattern Recognition (CVPR), 2019