Partial visual-semantic embedding: Fine-grained outfit image representation with massive volumes of tags via angular-based contrastive learning

Ryotaro Shimizu, Takuma Nakamura, Masayuki Goto

Abstract: Highlights•Model to extend the system for interpreting ambiguous fashion expressions is proposed.•Partial visual-semantic embedding for fine-grained learning for each part is proposed.•Part-by-part representations can be obtained via angular-based contrastive learning.•Full-body outfit images are embedded with rich fashion-specific ambiguous expressions.•Multiple unique evaluation experiments demonstrate the effectiveness of the proposal.