AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Towards Better Robustness against Common Corruptions for Unsupervised Domain Adaptation

Zhiqiang Gao, Kaizhu Huang, Rui Zhang, Dawei Liu, Jieming Ma

ICCV 2023 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Adversarial domain adaptation · 11 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add image discretization to DDAR (ablation)CDAN+MCC+DDAR · Office-31 corruptedmCE: 36.8 -> 30.5Improved
add adversarial perturbation to image discretization (ablation)CDAN+MCC+DDAR · Office-31 corruptedmCE: 33.8 -> 30.5Improved
add Gaussian random start for adversarial data (ablation)CDAN+MCC+DDAR · Office-31 corruptedmCE: 31.7 -> 30.5No clear effect
use Wasserstein regularization instead of KL or CDAN loss (ablation)DDAR with discriminator adversarial generation · Office-31 corruptedmCE: 38.9 -> 29.8Improved
generate adversarial data with discriminator instead of KL loss (ablation)DDAR · Office-31 corruptedmCE: 30.3 -> 29.8Improved
add DDAR to MCC+CDANMCC+CDAN · VisDA-2017mCE: 65.8 -> 55.6Improved
add DDAR to MCC on VisDA-2017MCC · VisDA-2017accuracy (%): 77.0 -> 66.8Hurt
add DDAR to MCC+CDAN on Office-31MCC+CDAN · Office-31accuracy (%): 89.6 -> 88.9Hurt
use pixel-space instead of discretized adversarial data (ablation)DDAR · Office-31 corruptedmCE: 30.5 -> 36.8Hurt
apply DDAR on top of Tent (epoch 1)CDAN+MCC+Tent · Office-31 corruptedmCE: 30.5 -> 25.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy