RGB-IR cross-modality person ReID based on teacher-student GAN model

• Teacher-Student model to minimize the different modality gap. Joint cycle-consistency GAN generate corresponding image pairs. Only use main backbone in test stage for efficiency. Numerous experiments conducted have proven effectiveness of proposed model. RGB-Infrared (RGB-IR) person re-identification (ReID) is a technology where system can automatically identify same appearing at parts video when light unavailable. The critical challenge this task cross-modality gap features under modalities. To solve challenge, we (TS-GAN) adopt domains and guide ReID backbone. (1) In order get RGB-IR pairs, Generative Adversarial Network (GAN) was used IR images. (2) kick-start training identities, Teacher module trained images, which then its Student counterpart training. (3) Likewise, better adapt domain enhance performance, three loss functions were used. Unlike other based models, only needs stage, making it more efficient resource-saving. showcase our model’s capability, did extensive on newly-released SYSU-MM01 RegDB Re-ID benchmark achieved superior performance state-of-the-art with 47.4% mAP 69.4% respectively.