Multiregion Bilinear Convolutional Neural Networks for Person Re-Identification

In this work we propose a new architecture for person re-identification. As task of re-identification is inherently associated with embedding learning non-rigid appearance description, our architecture is based on the deep convolutional network (Bilinear-CNN) that has been proposed recently fine-grained classification of highly non-rigid objects. While the last of the original Bilinear-CNN architecture completely removes the information from consideration by performing orderless pooling, we that a better embedding can be learned by performing bilinear pooling a more local way, where each pooling is confined to a predefined region. Our thus represents a compromise between traditional convolutional and bilinear CNNs and strikes a balance between rigid matching and ignoring spatial information. We perform the experimental validation of the new architecture on the three benchmark datasets (Market-1501, CUHK01, CUHK03), comparing it to that include Bilinear-CNN as well as prior art. The new architecture the baseline on all three datasets, while performing better than-of-the-art on two out of three. The code and the pretrained models of the can be found at https: //github. com/madkn/MultiregionBilinearCNN-ReId.