Unsupervised Tracklet Person Re-Identification

1 Mar 2019  ยท  Minxian Li, Xiatian Zhu, Shaogang Gong ยท

Most existing person re-identification (re-id) methods rely on supervised model learning on per-camera-pair manually labelled pairwise training data. This leads to poor scalability in a practical re-id deployment, due to the lack of exhaustive identity labelling of positive and negative image pairs for every camera-pair. In this work, we present an unsupervised re-id deep learning approach. It is capable of incrementally discovering and exploiting the underlying re-id discriminative information from automatically generated person tracklet data end-to-end. We formulate an Unsupervised Tracklet Association Learning (UTAL) framework. This is by jointly learning within-camera tracklet discrimination and cross-camera tracklet association in order to maximise the discovery of tracklet identity matching both within and across camera views. Extensive experiments demonstrate the superiority of the proposed model over the state-of-the-art unsupervised learning and domain adaptation person re-id methods on eight benchmarking datasets.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Person Re-Identification CUHK03 UTAL MAP 42.3 # 13
Rank-1 56.3 # 14
Person Re-Identification DukeMTMC-reID UTAL Rank-1 62.3 # 79
mAP 44.6 # 84
Person Re-Identification DukeTracklet UTAL Rank-1 43.8 # 1
Rank-20 76.5 # 1
Rank-5 62.8 # 1
mAP 36.6 # 1
Person Re-Identification iLIDS-VID UTAL Rank-1 35.1 # 10
Rank-20 83.8 # 7
Rank-5 59 # 7
Person Re-Identification Market-1501 UTAL Rank-1 69.2 # 103
mAP 46.2 # 109
Person Re-Identification MARS UTAL mAP 35.2 # 20
Rank-1 49.9 # 16
Rank-10 66.4 # 4
Rank-20 77.8 # 4
Person Re-Identification MSMT17 UTAL Rank-1 31.4 # 33
mAP 13.1 # 33
Person Re-Identification PRID2011 UTAL Rank-1 54.7 # 10
Rank-20 96.2 # 10
Rank-5 83.1 # 7

Methods


No methods listed for this paper. Add relevant methods here