SVDNet for Pedestrian Retrieval

This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (re-ID). We view the weight matrix $W$ as a set of weight vectors or basis. It is observed that the weight vectors are usually highly-correlated. This problem leads correlations among entries of the FC descriptor, and compromises the retrieval performance based on the Euclidean distance. To alleviate the correlation problem, this paper proposes to decorrelate the learned weight vectors using singular vector decomposition (SVD). Specifically, we design a novel training strategy with the "restraint and relaxation iteration" (RRI) scheme. We conduct experiment on the Market-1501, CUHK03, and Duke datasets, and show that RRI effectively reduces the correlation among the projection vectors and significantly improves the re-ID accuracy. On the Market-1501 dataset, for instance, rank-1 accuracy is improved from 55.2% to 80.5% for CaffeNet, and from 73.8% to 83.1% for ResNet-50.