Domain adaptation methods aim to improve object detection performance in the presence of distribution shifts. Multi-source domain adaptation enhances adaptation, generalization, and robustness. Existing methods focus on class-agnostic alignment, leading to challenges with unique object modal information. A recent prototype-based approach suffers from error accumulation due to noisy pseudo-labels. To address these limitations, an attention-based class-conditioned alignment scheme is proposed for multi-source domain adaptation. The method aligns instances of each object category across domains using an attention module and adversarial domain classifier. Experimental results show that the proposed method outperforms state-of-the-art methods and is robust to class imbalance.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Atif Belal,A... at arxiv.org 03-18-2024
https://arxiv.org/pdf/2403.09918.pdfDeeper Inquiries