AI Paper Atlas

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

Reliability in Semantic Segmentation: Can We Use Synthetic Data?

Thibaut Loiseau, Tuan-Hung Vu, Mickael Chen, Patrick Pérez, Matthieu Cord

ECCV 2024 · Detection and segmentation › Open-set and few-shot detection · approach: Amodal and open-world instance segmentation · 23 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
evaluate segmenters on synthetic OOD images instead of the Cityscapes validation set to predict real shift robustness40 pretrained segmenters · ACDC night/snowPearson correlation with real mIoUImproved
evaluate on synthetic data in small-gap domains (IDD, ACDC fog)40 pretrained segmenters · IDD, ACDC fogPearson correlationNo clear effect
use about 500 synthetic imagessynthetic robustness benchmark · robustness correlationPearson correlation stabilityImproved
swap SD1.5 generator for SDXL (ablation)synthetic benchmark generator · Fog and IndiaPearson correlation: 0.77 -> 0.89 (fog); 0.71 -> 0.93 (india)Improved
swap SD1.5 generator for SDXL (ablation)synthetic benchmark generator · Night and SnowPearson correlation: 0.85 -> 0.84 (night); 0.85 -> 0.82 (snow)No clear effect
recalibrate with temperature scaling on synthetic shift data40 pretrained segmenters · ACDC shifts and IDDECE improvement success rate: 72.5% on IDD, over 90% on ACDCImproved
calibrate with synthetic data instead of Cityscapes datasegmenters · shifted domainsECE improvementImproved
assess OOD detection with curated vs all inpainted images (ablation)synthetic OOD object benchmark · SMIYC RoadAnomaly21correlation with real scores: entropy PCC 0.94 for both; curated around 0.8 on anomaly metricsMixed
train RbA with synthetic inpainted OOD objectsRbA Swin-B · SMIYC RoadAnomaly21AUROC: 95.6 -> 97.3Improved
curate inpainted images for OOD training (ablation)RbA · SMIYC RoadAnomaly21AUROC: 97.2 (curated) vs 97.3 (all)No clear effect
use synthetic OOD training data instead of copy-pasted COCO objectsRbA Swin-L · SMIYC RoadAnomaly21AUROC: 98.1 (ours all) vs 98.2 (COCO)No clear effect
use per-class vs standard temperature scalingcalibration · synthetic calibrationECE improvementMixed

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