Learning Identity-Consistent Feature for Cross-Modality Person Re-Identification via Pixel and Feature Alignment

Figure 1

Framework of the proposed model. It consists of an image generation module (G), a joint discriminator module (D), and a feature extraction module (F). The G can generate fake IR images to mitigate the cross-modality variation, and the F can alleviate the intramodality variation. The F module contains ResNet-50 and BNG attention and MMM module. The BNG module can focus on channel and spatial information, and the MMM module can reduce modality differences.