AI Paper Atlas

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

Bridging Theory and Practice in Source-Free Domain Adaptation via Adversarial Proxy Perturbation

Dexuan Zhang, Qianyu Zhou, Thomas Westfechtel, Yusuke Mukuta, Tatsuya Harada

ECCV 2026 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Source-free domain adaptation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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 proxy perturbation and consistency losses (Ldis + Lce) (ablation)APP · Office-Home Ar->Claccuracy (%): 57.6 -> 61.6Improved
add adversarial loss Ladv (ablation)APP · DomainNet P->Saccuracy (%): 68.6 -> 72.2Improved
add self-supervised regularizers Lssl (ablation)APP · DomainNet P->Raccuracy (%): 79.7 -> 81.3Improved
add Lssl (ablation)APP · Office-Home Ar->Claccuracy (%): 60.2 -> 61.6Improved
add adversarial loss Ladv (ablation)APP · Office-Home Ar->Rwaccuracy (%): 81.9 -> 82.9Improved
co-train with CLIP via mutual distillation (APP -> APPM)APP · DomainNet (avg)accuracy (%): 73.8 -> 84.5Improved
APP vs strongest prior unimodal SFDAAPP · Office-Home (avg)accuracy (%): 74.1 -> 74.6Improved
APP vs strongest prior unimodal SFDAAPP · VisDA-Caccuracy (%): 89.1 -> 89.5Improved
APPM vs DIFO on open-set SFDAAPPM · Office-Home open-set SFDAaccuracy (%): 75.9 -> 85.2Improved
APPM vs DIFO on partial-set SFDAAPPM · Office-Home partial-set SFDAaccuracy (%): 85.6 -> 88.3Improved

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