Self and Channel Attention Network for Person Re-Identification

Recent research has shown promising results for person re-identification by focusing on several trends. One is designing efficient metric learning loss functions such as triplet loss family to learn the most discriminative representations. The other is learning local features by designing part based architectures to form an informative descriptor from semantically coherent parts. Some efforts adjust distant outliers to their most similar positions by using soft attention and learn the relationship between distant similar features. However, only a few prior efforts focus on channel-wise dependencies and learn non-local sharp similar part features directly for the degraded data in the person re-identification task. In this paper, we propose a novel Self and Channel Attention Network (SCAN) to model long-range dependencies between channels and feature maps. We add multiple classifiers to learn discriminative global features by using classification loss. Self Attention (SA) module and Channel Attention (CA) module are introduced to model non-local and channel-wise dependencies in the learned features. Spectral normalization is applied to the whole network to stabilize the training process. Experimental results on the person re-identification benchmarks show the proposed components achieve significant improvement with respect to the baseline.

I. Introduction

Person re-identification (re-id) is a task to retrieve the same person images from gallery sets of non overlapping cameras, given an image of a person of interest from another camera. The task of re-id is gaining increasing importance as it is an essential component of intelligent surveillance systems [1], [2]. The variations like illumination, human pose, view angle, resolution, occlusions, clothing and background in the images make re-id a very challenging task. With the advancement of deep learning and neural networks, ConvNets [3]–[4], [5], well designed for image classification tasks, are performing well in re- id as they provide impressive feature representations of person images. Due to their discriminative representation capability, they outperform the traditional handcrafted low-level features by a large margin. The difference between re- id and image classification tasks is that the training and testing classes (i.e. person identities) are different in re-id. Therefore, the re-id task requires more discriminative feature representations to distinguish unseen similar images.