AI Paper Atlas

Research map / Detection and segmentation / CNN features

Certifying the Full YOLO Pipeline: A Probabilistic Verification Approach

Zongxin Liu, Lijia Yu, Tao Lin, Zhiming Chi, Lijun Zhang

ICLR 2026 · Detection and segmentation › CNN features · approach: Box regression and loss design

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 ODPV instead of RCPN (10^6 samples) for IoU lower boundsYOLO verification · COCO val, eps=1/255, tau=0.5verification time: 563.5 -> 109.0 (YOLOv3-spp); 618.1 -> 147.2 (YOLO11x)Improved
use ODPV instead of RCPN for certified IoU lower boundsYOLO verification · COCO val, eps=1/255, tau=0.5mean abs. IoU gap to PGD: 0.54 -> 0.49 (YOLO11x)Improved
use ODPV instead of RCPN or DeepPAC for the same 98% guaranteePAC verification of YOLO · 98% guarantee at 98% confidencesamples required: 11,000,000 -> 37,000 (vs RCPN)Improved
increase perturbation radius from 1/255 to 2/255 (ablation)ODPV on YOLO11x · COCO val guarantee eval, tau=0.5true positive rate: 94.9 -> 85.2Hurt
increase perturbation radius from 1/255 to 2/255 (ablation)ODPV on YOLO11x · COCO val guarantee eval, tau=0.5certified robust accuracy: 98.9 -> 99.4No clear effect
add the Part 3 refinement step to the verification (ablation)ODPV on YOLO11x · COCO val guarantee eval, tau=0.5average bounds improvement: 0.10 (eps 1/255); 0.17 (eps 2/255)Improved

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