AI Paper Atlas

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

SCOD: From Heuristics to Theory

Vojtech Franc, Jakub Paplham, Daniel Prusa

ECCV 2024 · 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
replace the SIRC combination with the theoretically optimal linear combination of conditional risk and OOD/ID likelihood ratio (ablation)SCOD selector (plugin, clean-OOD LR) · ImageNet vs SSB_HardAuSRT: 13.24 -> 6.10Improved
combine MSP with the learned likelihood ratio through the linear score instead of using MSP aloneSCOD selector · ImageNet vs SSB_HardAuSRT: 17.47 -> 6.10Improved
use the corrected sigmoid instead of the standard sigmoid to learn the likelihood ratio (ablation)Linear double-score selector (POSCOD) · ImageNet vs SSB_HardAuSRT: 7.28 -> 6.37Improved
use the corrected sigmoid instead of the standard sigmoid for SIRC (ablation)SIRC selector · ImageNet vs iNaturalistAuSRT: 4.34 -> 8.02Hurt
replace SIRC with the linear strategy using POSCOD likelihood ratioSCOD selector · ImageNet, five OOD setsAuSRT: 14.27 -> 6.37 (SSB_Hard)Improved
use linear combination with contemporary OOD scores instead of SIRCSCOD selector · ImageNet, contemporary OOD scoresAuSRTMixed
vary relative cost alpha from 0.1 to 0.9 for Linear vs SIRC (ablation)POSCOD linear strategy · ImageNet, five OOD setsAuSRT: SSB_Hard alpha=0.9: 20.13 -> 9.97Improved
reduce OOD proportion in the training mixture from 0.5 to 0.1 (ablation)POSCOD · ImageNet vs SSB_HardAuSRT: 5.88 -> 17.20Hurt
tune strategy hyperparameters on test data (tuned vs plugin) (ablation)Linear and SIRC selectors · ImageNet vs SSB_HardAuSRT: Linear 6.10 -> 5.88; SIRC 13.24 -> 6.59Improved

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