AI Paper Atlas

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

Residual Pattern Learning for Pixel-Wise Out-of-Distribution Detection in Semantic Segmentation

Yuyuan Liu, Choubo Ding, Yu Tian, Guansong Pang, Vasileios Belagiannis, Ian Reid, Gustavo Carneiro

ICCV 2023 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 60 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 positive energy loss instead of entropy maximisation (ablation)RPL · SMIYC Obstacle (val)AuPRC: 80.85 -> 92.36Improved
positive energy loss instead of hinge energy loss (ablation)RPL · SMIYC Obstacle (val)AuPRC: 71.40 -> 92.36Improved
add dis-similarity regularisation (ablation)RPL+PE · Fishyscapes L&F (val)AuPRC: 56.85 -> 63.72Improved
add CoroCL (ablation)RPL+PE+DS · SMIYC Anomaly (val)AuPRC: 76.43 -> 88.55Improved
add CoroCL (ablation)RPL+PE+DS · RoadAnomaly (val)AuPRC: 63.02 -> 71.61Improved
add RPL as residual module added to frozen segmentation features instead of direct optimisation (ablation)RPL · Fishyscapes L&F (val)FPR: 17.18 -> 4.52Improved
add RPL to Meta-OoDMeta-OoD · Fishyscapes L&F (val)AuPRC: 41.31 -> 55.25Improved
add RPL to PEBALPEBAL · SMIYC Obstacle (val)AuPRC: 10.15 -> 84.23Improved
use single-layer projector instead of 2-layer in CoroCL (ablation)CoroCL · SMIYC Anomaly (val)AuPRC: 83.24 -> 88.55Improved
projector embedding depth 304 instead of 512 (ablation)CoroCL · SMIYC Anomaly and RoadAnomalyAuPRC: around 5% AuPRCImproved
use COCO-context samples also in the anchor set and contrastive pairs (ood inlier-ood inlier, ood inlier-outlier) (ablation)CoroCL · Fishyscapes and SMIYC (val)AuPRC: Static 92.46 -> 90.24; L&F 70.61 -> 60.72Hurt
freeze segmentation network with RPL instead of re-trainingRPL · Cityscapes mIoUinlier mIoU: PEBAL -1.27% mIoU, Meta-OoD -0.95% mIoU vs none for RPLImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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