Domain Adaptive Person Re-Identification via Camera Style Generation and Label Propagation

Abstract

Unsupervised domain adaptation in person reidentification resorts to labeled source data to promote the model training on target domain, facing the dilemmas caused by large domain shift and large camera variations. The non-overlapping labels challenge that source domain and target domain have entirely different persons further increases the re-identification difficulty. In this paper, we propose a novel algorithm to narrow such domain gaps. We derive a camera style adaptation framework to learn the style-based mappings between different camera views, from the target domain to the source domain, and then we can transfer the identity-based distribution from the source domain to the target domain on the camera level. To overcome the non-overlapping labels challenge and guide the person re-identification model to narrow the gap further, an efficient and effective soft-labeling method is proposed to mine the intrinsic local structure of the target domain through building the connection between GAN-translated source domain and the target domain. Experiment results conducted on real benchmark datasets indicate that our method gets state-of-the-art results.

Cross-Domain Person Re-Identification Based on Feature Fusion

“…There also exists style gaps in the target domain itself, namely intra-domain variations, which attracts scholars' attention [21], [49]. For example, Ren et al propose the CSGLP model simultaneously to learn the transfer mapping among different camera views and multiple domains, exploring StarGAN [5] as a camera style adaptation network [22]. Zhong et al propose the HHL model which uses StarGAN to transfer images from each camera to the others in target domain [45].…”

“…Generative Adversarial Networks (GAN) is used to realize the image style transfer of different datasets to alleviate the effect of cross-domain [9], [13]. However, the existing GAN-based models ignore complete expressions and occlusion of pedestrian characteristics, resulting in low accuracy in feature extraction [12], [22]. To address these issues, we introduce a cross domain model based on feature fusion (FFGAN) to fuse global, local and semantic features to extract more delicate pedestrian features.…”

Proxy Task Learning For Cross-Domain Person Re-Identification

“…Regarding feature-level, ARN [30] extracted the domain-invariant features through the encoding and decoding network to eliminate the domain differences. Some methods try to use the relationship between the source domain and the target domain as soft-label to guide training [31], [32]. CSCLP [31] constructed soft-labels by mining the semantic similarity between the target images and generated images.…”

Discriminative and Geometry-Aware Unsupervised Domain Adaptation

“…Since the target domain is unlabeled, these works rely on predicting pseudo-labeling (Kang et al, 2019) or computing prototype representations of source and target classes (Wang & Breckon, 2020), and then the target domain samples are classified by the prototype of the target domain classes during the training process. Structure-preserving methods (Ren et al, 2019;Xia & Ding, 2020) try to achieve the class-level transfer by matching the structure graphs across domains. However, as observed by Chen et al (2019), the discrim-inability may be decreased when the models only focus on enhancing transferability.…”