Multi-granularity cross attention network for person re-identification

Typical person re-identification (Re-ID) methods suffer from common challenges from body misalignment, occlusion issues, background perturbance, pose variations, and other aspects. In solving these problems, the combination of global features and local features makes the network pay attention to the global information and local information in the image. The attention mechanism is found to be effective, which aims to strengthen the salient information and suppress the irrelevant ones. To further enhance the contribution of global information to significant information, in this paper, we propose a multi-granularity cross attention (MGCA) network for person Re-ID. The key component of our framework is the multi-granularity cross attention module, where the attention module selectively aggregates the features of each location and extracts the weighted sum of the features of each location based on each pixel's contribution to significance. Thus, it obtains the global view of the image and the spatial correlation between any two positions. The related semantic features reinforce each other, further improving compactness and semantic consistency within the classes, gaining feature refinement and feature-pair alignment, respectively. Extensive experiments demonstrate that our method is comparable to the most advanced methods.