Improve Person Re-Identification with Part Awareness Learning
Houjing Huang,Wenjie Yang,Jinbin Lin,Guan Huang,Jiamiao Xu,Guoli Wang,Xiaotang Chen,Kaiqi Huang
DOI: https://doi.org/10.1109/tip.2020.3003442
IF: 10.6
2020-01-01
IEEE Transactions on Image Processing
Abstract:Person re-identification (ReID) aims to predict whether two images from different cameras belong to the same person. Due to low image quality and variance in view point and body pose, it remains a difficult task. To solve the task, a model is supposed to appropriately capture features that describe body regions for identification. With the simple intuition that explicitly incorporating ReID model with part awareness could be beneficial for learning a more discriminative feature space, we propose part segmentation as an assistant body perception task during the training of a ReID model. Specifically, we add a lightweight segmentation head to the backbone of ReID model during training, which is supervised with part labels. Note that our segmentation head is only introduced during training and that it does not change network input or the way of extracting ReID feature. Experiments show that part segmentation considerably improves the performance of ReID. Through quantitative and qualitative analyses, we further reveal that body part perception helps ReID model to capture a set of more diverse features from the body, with decreased similarity between part features and increased focus on different body regions. We experiment with various representative ReID models and achieve consistent improvement on several large-scale datasets including Market1501, CUHK03, DukeMTMC-reID and MSMT17. E.g. on MSMT17, our method increases Rank-1 Accuracy of GlobalPool-ResNet-50, PCB and MGN by 2.3%, 2.9% and 3.9%, respectively. Incorporated with MGN, our model achieves state-of-the-art performance, with Rank-1 Accuracy 95.8%, 78.8%, 90.0% and 84.0% on four datasets, respectively.