Yifei Wang

The success of vision-language models is primarily attributed to effective cross-modal alignment between vision and language. However, modality gaps persist even in well-aligned models and may be necessary for human perception, as evidenced by modality-specific phenomena such as visual texture an...