Self-Similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-Identification

Abstract

Domain adaptation in person re-identification (re-ID) has always been a challenging task. In this work, we explore how to harness the similar natural 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 the global body to local parts) of unlabeled samples to build multiple clusters from different views automatically. 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→Market1501) and 4.4% (Market1501→DukeMTMC) in mAP, respectively. Upon our SSG, we further introduce a clustering-guided semisupervised approach named SSG ++ to conduct the oneshot 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 .

Attribute Memory Transfer Network for Unsupervised Cross-Domain Person Re-Identification

“…In experiments, we employ several widely-recognized methods to compare with our AMTNet, which are all for unsupervised person re-ID task, and divided into two collections of (1) hand-crafted feature based methods: LOMO [20], Bow [48], UMDL [25], CAMEL [42]; (2) deep learning based unsupervised re-ID models: SPGAN [6], TJAIDL [34], SSG [9], ECN [53], MMCL [32].…”

“…For another aspect, several researches [8], [19], [30], [45] introduce a dependable pseudo identity labels generator to assist the training in unlabeled target domain. Specifically, the works [8], [30], [45] utilize the density-based clustering algorithm [3], [7] for label estimation, which has improved the pseudo-label reliability. In addition, the works [8], [45] also employ part division as the supplemental feature to obtain satisfactory performance.…”

“…Specifically, the works [8], [30], [45] utilize the density-based clustering algorithm [3], [7] for label estimation, which has improved the pseudo-label reliability. In addition, the works [8], [45] also employ part division as the supplemental feature to obtain satisfactory performance. In practical, these methods still obtain limited performance because they neglect the detailed local information in unlabeled target data.…”

“…The second aspect of the obstacles is the intra-domain variations, especially for target domain due to the lacking of identity labels. Aiming at this aspect, several recent approaches attempt to explore the underlying correlations in completely unlabeled target domain [8], [46], [53]. To utilize the invariant information in the cross-camera pedestrian images, ECN [53] investigates an exemplar memory to store features of the target domain and accommodate the properties by solely measuring the exemplar-level similarity which is often fragile and can be misled easily, since it ignores the fine-grained information between similar but different identities as well as the global distribution of the pedestrian features with large variances in the same identity; SSG [8] employs a self-similarity grouping method to exploit the potential similarity from the global body to local parts of unlabeled samples to build multiple clusters from different views automatically; MMN [46] integrates a multi-level memory network to maintain the upto-date representation of the entire dataset, discovering the multi-level memory, and domain-level memory to capture the fine-grained differences between images and the global structures for the holistic target domain.…”

Receptive Multi-Granularity Representation for Person Re-Identification

“…learning on the dataset including identities with only one labeled example along with many unlabeled examples. Fu et al [55] focus on local-based unsupervised feature learning, and inherit the semi-supervised setting with self-grouping strategies.…”