Clothes-Changing Image Generation Based on Attention for Person Re-identification
Cheng Chun Tang, Jie Guo · 2020 5th International Conference on Mechanical, Control and Computer Engineering (ICMCCE) · 2020
Person re-identification is an important issue in the field of video surveillance security. However, the differences of pedestrians under different cameras bring challenges. The existing person re-identification methods mainly focus on the person's posture and feature extraction, but do not pay much attention to the high-frequency clothes changing for one person. Therefore, the study of person clothes-changing images is of great help to person re-identification. In this paper, we introduce a network which fuses feature separation and self-attention jointly in a generative adversarial network for image generation of person clothes. We evaluate this model based on self-attention on a benchmark dataset: Market-1501. The experiment results validate the advances of attention mechanism by comparing with state-of-the-art baselines. Based on the generated datasets, we introduce a network containing channel attention and spatial attention which focus on local regions of clothes changes in a convolutional neural network for image recognition. We evaluate this network on our generative dataset. The experiment results show that joint attention network outperforms the baseline.