AI Paper Atlas

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

Reliability in Semantic Segmentation: Are We on the Right Track?

Pau de Jorge, Riccardo Volpi, Philip H.S. Torr, Grégory Rogez

CVPR 2023 · Detection and segmentation › Open-set and few-shot detection · approach: Interactive and referring segmentation · 24 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
Use Local Temperature Scaling (LTS) instead of temperature scaling (TS)Segmentation models calibrated on Cityscapes · ACDCECEImproved
Calibrate with LTSSegmentation models · ACDCPRRImproved
Calibrate with clustering TS or LTSSegmentation models · IDD and ACDCAUROCImproved
Use clustering calibration with Cityscapes-only examplesSegmentation models · IDD and ACDCECENo clear effect
Add naive data augmentation to the Cityscapes calibration set (ablation)Clustering TS and LTS · OOD calibrationECENo clear effect
Use all datasets for calibration rather than Cityscapes only (ablation)Clustering TS · IDD and ACDCECEImproved
Increase the number of clustering groups (ablation)Adaptive clustering TS · Calibration set and OOD test datasetsECEMixed
Use per-class instead of per-image clustering (ablation)Adaptive clustering calibration · Segmentation calibrationECENo clear effect
Use only image input instead of only logits as LTS input (ablation)LTS · ACDCECEImproved
Use only logits rather than both image and logits as LTS input (ablation)LTS · Cityscapes (CS)ECEImproved
Use max probability rather than negative entropy as confidence score (ablation)Misclassification and OOD detection · Cityscapes, IDD, and ACDCDetection performanceNo clear effect
Increase sampled calibration pixels to more than 10k per image (ablation)Calibration metric estimation · Semantic segmentation calibrationECE stabilityNo clear effect

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