AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Image-based Outlier Synthesis With Training Data

Sudarshan Regmi

CVPR 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based OOD detection · 4 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 gradient to ID inputs instead of subtracting it for outlier synthesis (ablation)ASCOOD · Car (fine-grained)FPR@95: 60.20 -> 40.76Improved
add gradient (vs pixel shuffling) for outlier synthesis (ablation)ASCOOD · Aircraft (fine-grained)FPR@95: 47.98 -> 47.94No clear effect
use z-score instead of L2 normalization of features (ablation)ASCOOD · Car (fine-grained)FPR@95: 51.86 -> 40.76Improved
use z-score instead of L2 normalization (ablation)ASCOOD · CIFAR-100FPR@95: 40.81 -> 29.90Improved
use KL-to-uniform loss instead of energy loss for outlier uncertainty (ablation)ASCOOD · Aircraft (fine-grained)FPR@95: 58.70 -> 47.94Improved
standardize features rather than raw features with KL loss (ablation)ASCOOD · Aircraft (fine-grained)FPR@95: 55.21 -> 47.94Improved
use i-ODIN (perturb only significant channels) instead of ODIN (ablation)ODIN · CIFAR-10 vs CIFAR-100FPR@95: 77.00 -> 61.33Improved
use i-ODIN instead of ODIN on two-class benchmarks (ablation)ODIN · Waterbirds and CelebAFPR@95No clear effect
train with ASCOOD vs baselineResNet · ID accuracyaccuracy: 76.63 vs 77.25 (CIFAR-100)No clear effect
ASCOOD vs external-data OE/MixOEASCOOD · Waterbirds (spurious)FPR@95: 33.26 -> 11.37Improved

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