Browse Teach Curate

BatStyler: Multi-Category Style Generation for Source-Free Domain Generalization

Improves source-free domain generalization in the multi-domain, multi-category regime via a Coarse Semantic Generation module that prevents style-diversity space from collapsing as categories grow and a Uniform Style Generation module that produces uniformly distributed styles trained in parallel, raising both diversity and efficiency of synthetic styles and surpassing SOTA on multi-category datasets.

KITT Text

BatStyler: Advancing Multi-category Style Generation for Source-free Domain Generalization Xiusheng

BatStyler tackles source-free domain generalization in the hard multi-domain, multi-category setting. It comprises a Coarse Semantic Generation module that extracts coarse-grained semantics to keep t…