Reproduce “Vision transformers in domain adaptation and domain generalization: a study of robustness” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shadi Alijani, Jamil Fayyad, Homayoun Najjaran
This review systematically categorizes and analyzes existing research on Vision Transformers (ViTs) in Domain Adaptation and Domain Generalization. It establishes that ViTs exhibit superior robustness to distribution shifts compared to CNNs, applying effectively across diverse tasks like semantic segmentation, action recognition, face analysis, and medical imaging.
Key insights
- Vision Transformers (ViTs) leverage dynamic weight computation and multi-head self-attention to capture global features, offering superior robustness to distribution shifts compared to CNNs.
- ViTs exhibit a reduced texture bias, favoring shape recognition, which contributes to their enhanced generalization capabilities across varied domains.
- Existing research often employs hybrid methods, combining multiple adaptation and generalization strategies to achieve more robust performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.