A Lightweight Model Based on Co-segmentation Attention for Occluded Person Re-identification

Occlusion generally exists in application scenarios of person re-identification, especially in crowded situations. It is challenging to solve the occlusion problem due to the differences in the size, shape and color of the occlusion. Although pose information is helpful in solving the occlusion problem, it usually takes much time and computing resources, and its accuracy is still limited at present. To address this issue, we propose a lightweight model that uses co-segmentation attention and local features to solve the occlusion problem by attention mechanism. Experimental results on three reported datasets show that the performance of our proposed model surpasses most existing methods and is competitive with the most state-of-the-art method.