AI Paper Atlas

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

Unsupervised Accuracy Estimation of Deep Visual Models using Domain-Adaptive Adversarial Perturbation without Source Samples

JoonHo Lee, Jae Oh Woo, Hankyu Moon, Kwonho Lee

ICCV 2023 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Geometric alignment and synthetic-to-real adaptation · 3 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 adjusted predictive uncertainty scaling (C_adj_unc) to uniform epsilon scaling (ablation)SF-DAP (ADV) · Office-31MAE (%): 5.76 -> 4.43Improved
uniformly scale epsilon by class complexity and density for all instances (ablation)SF-DAP (ADV) · Office-31MAE (%): 4.72 -> 5.76Hurt
add class complexity factor C_cls with C_adj_unc (ablation)SF-DAP (AAP) · Office-31MAE (%): 4.70 -> 3.05Improved
add data density factor C_den on top of C_cls and C_adj_unc (ablation)SF-DAP (AAP) · Office-31MAE (%): 3.05 -> 2.51Improved
replace ADV with adaptive adversarial perturbation (AAP) (ablation)SF-DAP · Overall macro average (63 settings)MAE (%): 4.86 -> 2.95Improved
apply AAP directly to source model without source-free UDA (ablation)SF-DAP (AAP) · Digits and Office-31average MAE: 3.0% increase in average MAEHurt
use SF-DAP (AAP) instead of RM under synthetic corruption shiftCIFAR-10-C / CIFAR-100-C · CIFAR-10-CMAE (%): RM 2.17 vs AAP 3.06Hurt
use source-free SF-DAP (AAP) instead of source-access baselines on natural shiftDigits, Office-31, VisDA · VisDAMAE (%): RM 4.41 -> AAP 1.73Improved
use PAFA instead of SHOT-IM for source-free UDA (ablation)source-free UDA · M->Starget accuracy (%): 11.0 -> 83.5Improved
use PAFA instead of FAUST (ablation)source-free UDA · UDA averagetarget accuracy (%): 94.6 -> 93.8No clear effect

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