Unified Modality Separation for Unsupervised Domain Adaptation
The author introduces a Unified Modality Separation framework for unsupervised domain adaptation, emphasizing the importance of balancing modality-specific information. The approach aims to disentangle CLIP's features into distinct language-associated and vision-associated components.
