Fully Unsupervised Person Re-identification viaSelective Contrastive Learning
This addresses the problem of reducing manual annotation costs for adapting person re-identification systems to new conditions, though it is incremental in improving unsupervised methods.
The paper tackles unsupervised person re-identification by proposing a selective contrastive learning framework that uses multiple positives and adaptively sampled negatives to learn identity-discriminative representations, achieving state-of-the-art results.
Person re-identification (ReID) aims at searching the same identity person among images captured by various cameras. Unsupervised person ReID attracts a lot of attention recently, due to it works without intensive manual annotation and thus shows great potential of adapting to new conditions. Representation learning plays a critical role in unsupervised person ReID. In this work, we propose a novel selective contrastive learning framework for unsupervised feature learning. Specifically, different from traditional contrastive learning strategies, we propose to use multiple positives and adaptively sampled negatives for defining the contrastive loss, enabling to learn a feature embedding model with stronger identity discriminative representation. Moreover, we propose to jointly leverage global and local features to construct three dynamic dictionaries, among which the global and local memory banks are used for pairwise similarity computation and the mixture memory bank are used for contrastive loss definition. Experimental results demonstrate the superiority of our method in unsupervised person ReID compared with the state-of-the-arts.