Crossmodality Person Reidentification Based on Global and Local Alignment

Figure 1

Our proposed framework (AGF) consists of a AlignGAN module (), a crossmodality paired-images generation module (), and a feature alignment module (). A generates fake IR images from RGB images to reduce the crossmodality gap, and then we transfer it to the framework as input images together with all the images in the SYSU-MM01 dataset. disentangles images to modality-specific and modality-invariant features and then decodes from the exchanged features. uses an encoder whose weights are shared with modality-invariant encoder to perform set-level alignment and then further performs instance-level alignment by minimizing distance between each pair images.