Universal domain adaptation from multiple black-box sources

Open Webpage

Yunyun Wang, Jian Mao, Cong Zou, Xinyang Kong

Published: 2024, Last Modified: 06 Nov 2025Image Vis. Comput. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Uni-MBDA is proposed for multiple black-box sources with label shift from target.•Um2B is proposed for Uni-MBDA with dual-attention on source domains and classes.•Empirical results show the effectiveness of Um2B for Uni-MBDA.