AI Paper Atlas

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

DINE: Domain Adaptation From Single and Multiple Black-Box Predictors

Jian Liang, Dapeng Hu, Jiashi Feng, Ran He

CVPR 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Feature alignment for unsupervised adaptation · 89 citations (OpenAlex)

Check 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 mutual information regularization during distillation and fine-tuning (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 63.6±3.5 -> 72.2±1.8Improved
add MixUp interpolation consistency regularization (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 72.0±1.7 -> 72.2±1.8Improved
add the second fine-tuning stage (ablation)DINE multi-source · Office multi-source closed-set UDA, ResNet-101 source (Avg.)accuracy: 92.1 -> 93.4Improved
use adaptive label smoothing with top-1 confidence rather than hard one-hot labels (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 67.3±1.7 -> 72.2±1.8Improved
use adaptive label smoothing with top-1 confidence rather than vanilla label smoothing (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 68.4±1.8 -> 72.2±1.8Improved
retain top 3 classes instead of using top 1 in adaptive label smoothing (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 72.2±1.8 -> 72.9±2.0Improved
use the full source prediction vector instead of top-1 adaptive smoothing (ablation)DINE · Office-Home partial-set UDA, ViT source → ResNet-50 target (Avg.)accuracy: 83.0 -> 82.1Hurt
use full source prediction vectors instead of top-1 adaptive smoothing (ablation)DINE · Office-Home closed-set UDA, D→Aaccuracy: 72.2±1.8 -> 72.9±2.5Mixed

Later papers that compared with this method

7 later accepted papers report DINE as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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