AI Paper Atlas

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

PixOOD: Pixel-Level Out-of-Distribution Detection

Tomas Vojir, Jan Sochman, Jiri Matas

ECCV 2024 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based OOD detection · 13 citations (OpenAlex)

Read the paper

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 multi-component class modelling with soft-to-hard condensation instead of a single mean (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 88.14 -> 94.80Improved
use soft-to-hard condensation instead of soft K-Means (K=200) (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 93.01 -> 94.11Improved
increase condensation budget K from 200 to 1000 (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 94.11 -> 94.80Improved
use an MLP classifier instead of a linear probe (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 92.36 -> 94.80Improved
up-sample feature map by 7 instead of 1 (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 92.14 -> 94.80Improved
require patch label percentage above 90% instead of 50% (ablation)PixOOD · Road anomaly (RA+FS+RO mean)mean AP: 94.22 -> 94.80Improved
add DaCUP reconstruction module as a second OOD head (25% weight) (ablation)PixOOD · Road AnomalyAP: 96.39 -> 98.66Improved
add DRAEM as specialised module to PixOOD (ablation)PixOOD · MVTec ADAUPRO: 91.3 -> 95.3Improved
PixOOD without OOD training data versus methods with OOD data on SMIYC Anomaly TrackPixOOD · SMIYC Anomaly TrackAP: 68.88 vs 94.46 (RbA with OOD data)Hurt
use PixOOD generic unified model on LaRS vs top leaderboardPixOOD · LaRSQ: 73.9 vs 78.1Mixed

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