We are delighted to announce that one paper from the Perceuptual AI Lab (PAI Lab, Prof. Chanho Eom) has been accepted to Journal of Visual Communication and Image Representation (JVCI).

Title: 

Modality-Specific Expert Guiding for Visible-Infrared Person ReIdentification

Authors:

Xinyu Guo, Xiaobin Liu, Zishao Qiao, Nan Hu, Jianing Li, Chanho Eom, and Jing Yuan

Abstract:

The Visible-Infrared person Re-Identification (VI-ReID) task requires optimizing feature distance both within and across different modalities, making it a more challenging multimodal learning task compared to the previous visible ReID task. Existing methods commonly optimize intra-modality and inter-modality feature distance on a single model simultaneously, leading to a complicated multimodal learning task and hindering the adequate optimization in the shared feature space. To handle this issue, this paper proposes a modality-Specific Expert Guiding (SEG) method for the VI-ReID task. Specifically, SEG first independently learns two modality-specific person ReID models for visible and infrared modalities as expert models, respectively. As each expert model focuses on feature optimization within the corresponding modality, these two experts are able to acquire stronger discrimination capacity for person images in each corresponding modality, hence could provide accurate modality-specific supervision on feature distance optimization for the cross-modality joint training of the model. The proposed SEG is a new multimodal learning paradigm for the VI-ReID task to disentangle the feature optimization within and across modalities, which is able to effectively optimize the VI-ReID model with cross-modality person images. Experiments on three widely-used VI-ReID datasets, i.e., SYSU-MM01, RegDB, and LLCM, demonstrate the superior performance of our SEG over existing methods.