Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification

ICCV 2019  ·  Yang Fu, Yunchao Wei, Guanshuo Wang, Yuqian Zhou, Honghui Shi, Thomas Huang ·

Domain adaptation in person re-identification (re-ID) has always been a challenging task. In this work, we explore how to harness the natural similar characteristics existing in the samples from the target domain for learning to conduct person re-ID in an unsupervised manner... Concretely, we propose a Self-similarity Grouping (SSG) approach, which exploits the potential similarity (from global body to local parts) of unlabeled samples to automatically build multiple clusters from different views. These independent clusters are then assigned with labels, which serve as the pseudo identities to supervise the training process. We repeatedly and alternatively conduct such a grouping and training process until the model is stable. Despite the apparent simplify, our SSG outperforms the state-of-the-arts by more than 4.6% (DukeMTMC to Market1501) and 4.4% (Market1501 to DukeMTMC) in mAP, respectively. Upon our SSG, we further introduce a clustering-guided semisupervised approach named SSG ++ to conduct the one-shot domain adaption in an open set setting (i.e. the number of independent identities from the target domain is unknown). Without spending much effort on labeling, our SSG ++ can further promote the mAP upon SSG by 10.7% and 6.9%, respectively. Our Code is available at: https://github.com/OasisYang/SSG . read more

PDF Abstract ICCV 2019 PDF ICCV 2019 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Unsupervised Person Re-Identification DukeMTMC-reID Self-Similarity Grouping (one shot) Rank-1 72.4 # 1
Rank-10 87.7 # 1
Rank-5 84 # 1
MAP 55.9 # 1
Unsupervised Person Re-Identification DukeMTMC-reID->MSMT17 Self-Similarity Grouping (one shot) mAP 23.6 # 2
Rank-1 43.6 # 2
Rank-10 61.8 # 1
Unsupervised Domain Adaptation Duke to Market SSG mAP 58.3 # 10
rank-1 80.0 # 11
rank-5 90.0 # 9
rank-10 92.4 # 10
Unsupervised Domain Adaptation Duke to MSMT SSG mAP 13.3 # 7
rank-1 32.2 # 7
rank-5 - # 8
rank-10 51.2 # 7
Unsupervised Person Re-Identification Market-1501 Self-Similarity Grouping (one shot) Rank-1 87.5 # 1
MAP 71.5 # 1
Rank-10 96.8 # 1
Rank-5 95.2 # 1
Unsupervised Person Re-Identification Market-1501->MSMT17 Self-Similarity Grouping (one shot) mAP 11.8 # 3
Rank-1 27.6 # 2
Rank-10 45.7 # 1
Unsupervised Domain Adaptation Market to Duke SSG mAP 53.4 # 10
rank-1 73.0 # 7
rank-5 80.6 # 8
rank-10 83.2 # 10
Unsupervised Domain Adaptation Market to MSMT SSG mAP 13.2 # 8
rank-1 31.6 # 8
rank-5 - # 9
rank-10 49.6 # 8

Methods


No methods listed for this paper. Add relevant methods here