Learning Feature Fusion for Unsupervised Domain Adaptive Person Re-identification

Abstract

Unsupervised domain adaptive (UDA) person reidentification (ReID) has gained increasing attention for its effectiveness on the target domain without manual annotations. Most fine-tuning based UDA person ReID methods focus on encoding global features for pseudo labels generation, neglecting the local feature that can provide for the fine-grained information.To handle this issue, we propose a Learning Feature Fusion (LF 2 ) framework for adaptively learning to fuse global and local features to obtain a more comprehensive fusion feature representation. Specifically, we first pre-train our model within a source domain, then fine-tune the model on unlabeled target domain based on the teacher-student training strategy. The average weighting teacher network is designed to encode global features, while the student network updating at each iteration is responsible for fine-grained local features. By fusing these multiview features, multi-level clustering is adopted to generate diverse pseudo labels. In particular, a learnable Fusion Module (F M ) for giving prominence to fine-grained local information within the global feature is also proposed to avoid obscure learning of multiple pseudo labels. Experiments show that our proposed LF 2 framework outperforms the state-of-the-art with 73.5% mAP and 83.7% Rank1 on Market1501 to DukeMTMC-ReID, and achieves 83.2% mAP and 92.8% Rank1 on DukeMTMC-ReID to Market1501.

Dual Pseudo Label Refinement for Unsupervised Domain Adaptive Person Re-Identification

“…Numerous remarkable methods have been proposed to train the model robustly with noisy labels, which can be grouped into noise-robust loss designing [43], [44], [45], [46], loss adjustment [12], [47], [48], [49], [50], [51] and pseudo label refinement [52], [54], [55]. Noise-robust loss designing methods define robust loss functions combating noisy lables.…”

“…UNRN [14] evaluates reliability of pseudo labels by estimating uncertainty in global feature space. The LF 2 [55] is a learnable feature fusion framework embeded in the mean teacher and student networks. The global and local features are fused together in a learnable fusion module to obtain a more comprehensive feature representation.…”