Deep Association Learning for Unsupervised Video Person Re-identification

Deep learning methods have started to dominate the research progress of-based person re-identification (re-id). However, existing methods mostly supervised learning, which requires exhaustive manual efforts for cross-view pairwise data. Therefore, they severely lack scalability practicality in real-world video surveillance applications. In this work, address the video person re-id task, we formulate a novel Deep Association (DAL) scheme, the first end-to-end deep learning method using none of identity labels in model initialisation and training. DAL learns a deep-id matching model by jointly optimising two margin-based association losses an end-to-end manner, which effectively constrains the association of each to the best-matched intra-camera representation and cross-camera. Existing standard CNNs can be readily employed within our DAL. Experiment results demonstrate that our proposed DAL significantly current state-of-the-art unsupervised video person re-id methods on benchmarks: PRID 2011, iLIDS-VID and MARS.