AI Paper Atlas

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

Data Invariants to Understand Unsupervised Out-of-Distribution Detection

Lars Doorenbos, Raphael Sznitman, Pablo Márquez-Neila

ECCV 2022 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 7 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
use MahaAD instead of MSCL (best prior method) as the detectorU-OOD detectors · U-OOD benchmark, task meanAUC: 84.1 -> 86.5Improved
swap ResNet-101 for EfficientNet-b4 as MahaAD backbone (ablation)MahaAD · U-OOD benchmark, task meanAUC: 86.5 -> 89.3Improved
swap ResNet-101 for EfficientNet-b4 as MahaAD backbone (ablation)MahaAD · shift-high-resAUC: 78.6 -> 83.8Improved
swap ResNet-101 for EfficientNet-b4 as MahaAD backbone (ablation)MahaAD · uni-anoAUC: 91.3 -> 94.4Improved
use MahaAD instead of SSD on low-resolution domain shiftU-OOD detectors · shift-low-res (CIFAR10:SVHN)AUC: 99.6 vs 94.3Hurt
use MahaAD instead of CFlow on anomaly detectionU-OOD detectors · uni-ano (MVTec)AUC: 95.7 vs 91.3Hurt
use MahaAD instead of MSCL on one-vs-rest CIFARU-OOD detectors · uni-classAUC: 96.3 vs 92.4Hurt
use only the lowest-variance principal components (invariants) in the Mahalanobis score (ablation)MahaAD · four U-OOD experiments, ResNet50AUC: converges at about 3% of variance explained vs over 80% for high-variance componentsImproved
use MahaAD instead of training-based detectors for training timeMahaAD · CIFAR10:SVHN training timetraining time: about 90 seconds vs more than 12 hours for SSDImproved
increase number of CIFAR10 training classes, effect on SVHN/MNIST detection (ablation)MahaAD · CIFAR10 multi-class trainingAUCNo clear effect
increase number of CIFAR10 training classes, effect on SVHN/MNIST detection (ablation)DN2 and MSCL · CIFAR10 multi-class trainingAUCHurt
add shrinkage to covariance estimate (ablation)MahaAD · U-OOD benchmarkAUCNo 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