AI Paper Atlas

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

Visualizing Adapted Knowledge in Domain Transfer

Yunzhong Hou, Liang Zheng

CVPR 2021 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Feature alignment for unsupervised adaptation · 65 citations (OpenAlex)

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
Generate source-style images from target images using the two models as guidanceSFIT · Digits, Office-31, and VisDArecognition accuracy on generated images: 26.5% to 0.2%; 7.6% to 0.7%; 25.2% to 0.3%; 8.0% to 1.7%; 33.9% to 6.9%Improved
Fine-tune the target model using generated-image cues and target imagesTarget model · Office-31accuracy: 0.4%Improved
Fine-tune the target model using generated-image cues and target imagesTarget model · VisDAaccuracy: 0.7%Improved
Add the knowledge distillation loss (ablation)SFIT · VisDAaccuracy: 1.1%Improved
Add the channel-wise relationship preserving loss (ablation)SFIT · VisDAaccuracy: 2.6%Improved
Replace channel-wise relationship preserving loss with traditional style loss (ablation)SFIT · VisDAaccuracy: 4.8%Hurt
Replace channel-wise relationship preserving loss with batch-wise relationship loss (ablation)SFIT · VisDAaccuracy: 71.2; 71.2No clear effect
Replace channel-wise relationship preserving loss with pixel-wise relationship loss (ablation)SFIT · VisDAaccuracy: 70.9; 71.2No clear effect
Use transparent-filter initialization rather than the original target images (ablation)SFIT generator · VisDAaccuracy: 46.8 -> 44.9Hurt

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