AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

PartImageNet++ Dataset: Scaling up Part-based Models for Robust Recognition

Xiao Li, Yining Liu, Na Dong, Sitian Qin, Xiaolin Hu

ECCV 2024 · Adversarial robustness and security › Certified and empirical robustness · approach: Robust overfitting and regularization · 7 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 multi-scale part supervision (MPM) to adversarially trained ResNet-50-GeluResNet-50-Gelu with AT · ImageNet-1K val, average over four attacksaverage robust accuracy: 19.3 -> 20.8Improved
add MPM part supervision without ATResNet-50-Gelu · ImageNet-Caverage accuracy: 40.8 -> 42.1Improved
add MPM part supervision with ATResNet-50-Gelu AT · OOD datasets averageaverage accuracy: 24.6 -> 25.2Improved
add MPM part supervision without ATResNet-50-Gelu · OOD datasets averageaverage accuracy: 23.3 -> 23.6No clear effect
add pseudo part labels for unannotated images (ablation)MPM · ImageNet-1K vall_inf robust accuracy: 33.0 -> 39.1Improved
add category filter in pseudo-label generation (ablation)Mask R-CNN pseudo labels · PIN++ part segmentation valAP: 37.2 -> 40.7Improved
replace two-stage part model with MPM bypass layers (ablation)part-based model · ImageNet-1K vall_inf robust accuracy: 38.7 -> 39.1Improved
supervise three scales instead of one (ablation)MPM · ImageNet-1K vall_inf (8/255) robust accuracy: 8.8 -> 13.6Improved
add top-down connections (ablation)MPM SV3 · ImageNet-1K vall_inf (8/255) robust accuracy: 11.7 -> 13.6Improved
use part masks instead of object masks for supervision (ablation)MPM · ImageNet-1K valaverage accuracy: 22.3 -> 23.2Improved
vary lambda above zero (ablation)MPM · Tiny-PartImageNet++adversarial robustnessNo clear effect
initialize detector backbone with MPM instead of vanillaFaster R-CNN · MS-COCOclean AP50: 49.8 -> 50.3Improved

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