Single Camera Person Re-identification with Self-paced Joint Learning

Abstract Existing re-identification (re-ID) methods rely on a large number of cross-camera identity tags for training, and the data annotation process is tedious time-consuming, resulting in difficult deployment real-world re-ID applications. To overcome this problem, we focus single camera training (SCT) setting, where each annotated camera. Since there no across cameras, it takes much less time acquisition, enables fast new environments. address SCT re-ID, proposed joint comparison learning framework split into three parts, single-camera labeled data, pseudo unlabeled instances. In framework, iteratively (1) train network dynamically update memory to store types (2) assign pseudo-labels images using clustering algorithm. model phase, jointly CNN model, method can continuously advantages both labeled, or images. Extensive experiments are conducted widely adopted datasets, including Market1501-SCT MSMT17-SCT, show superiority our SCT. Specifically, mAP significantly outperforms state-of-the-art by 42.6% 30.1%, respectively.