In the study of Universal Semi-Supervised Domain Adaptation (UniSSDA), a new strategy is proposed to mitigate common-class bias by refining pseudo-labels. Existing methods are shown to be vulnerable to this bias, affecting the adaptation performance in challenging settings. The proposed strategy effectively improves target accuracy without sacrificing common class accuracy across various datasets and models. It establishes a new baseline for future research in this area.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Wenyu Zhang,... at arxiv.org 03-19-2024
https://arxiv.org/pdf/2403.11234.pdfDeeper Inquiries