AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Road Anomaly Detection by Partial Image Reconstruction With Segmentation Coupling

Tomas Vojir, Tomáš Šipka, Rahaf Aljundi, Nikolay Chumerin, Daniel Olmeda Reino, Jiri Matas

ICCV 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Industrial anomaly detection · 96 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 the SSIM reconstruction module to the segmentation network (ablation)JSR-Net · Average over 4 datasetsAP (%): 31.9 -> 78.9Improved
add trained segmentation coupling to the untrained product (ablation)JSR-Net · Average over 4 datasetsAP (%): 78.9 -> 79.1No clear effect
add auxiliary SSIM reconstruction loss L_R (ablation)JSR-Net · Average over 4 datasetsAP (%): 79.1 -> 82.9Improved
use SSIM instead of L2 as reconstruction error (ablation)JSR-Net · Average over 4 datasetsAP (%): 69.5 -> 82.9Improved
add synthetic anomaly polygon augmentation (ablation)JSR-Net · Average over 4 datasetsAP (%): 60.9 -> 82.9Improved
use SSIM instead of L2 reconstruction error (ablation)JSR-Net · Average over 4 datasetsFPR95 (%): 10.2 -> 5.1Improved
add the synthetic augmentation (ablation)JSR-Net · Average over 4 datasetsFPR95 (%): 11.5 -> 5.1Improved
switch backbone from Mobilenet v2 to ResNet-101 checkp1 (ablation)JSR-Net · Average over 4 datasetsAP (%): 78.7 -> 82.9Improved
switch backbone from Xception to ResNet-101 checkp1 (ablation)JSR-Net · Average over 4 datasetsAP (%): 82.1 -> 82.9No clear effect
use stronger segmentation checkpoint (51.6 -> 66.1 mIoU) (ablation)JSR-Net ResNet-101 · Average over 4 datasetsAP (%): 82.9 -> 83.7No clear effect
replace Resynthesis-PSPNet with JSR-Net checkp2Road Anomaly detection · Road AnomalyAP (%): 76.4 -> 94.4Improved
replace Resynthesis-PSPNet with JSR-Net checkp1Road anomaly detection · Lost and FoundFPR95 (%): 43.1 -> 3.6Improved

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