AI Paper Atlas

Research map / Image generation / Image restoration

Asynchronous Event Error-Minimizing Noise for Safeguarding Event Dataset

Ruofei Wang, Peiqi Duan, Boxin Shi, Renjie Wan

ICCV 2025 · Image generation › Image restoration · approach: Raw and low-light noise modeling · 1 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 similarity loss to surrogate training (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 3.85 -> 0.52Improved
add similarity loss to surrogate training (ablation)UEvs sample-wise noise · N-Caltech101 (ViT-B)test accuracy (%): 37.51 -> 1.09Improved
mix class-wise and sample-wise noise by random selection (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 0.52 -> 5.17No clear effect
mix class-wise and sample-wise noise by addition (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 0.52 -> 5.34Mixed
hacker uses data augmentation (shift, crop, flip, event drop) (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 0.52 -> 13.04Hurt
generate noise with FGSM instead of PGD (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 0.52 -> 27.74Hurt
raise projection tau from 3/4 to 7/8 (ablation)UEvs sample-wise noise · N-Caltech101 (ResNet18)test accuracy (%): 0.52 -> 4.88Hurt
add noise projection to sparsify into {-0.5,0,0.5} (ablation)UEvs sample-wise noise · event stream compatibilityreconstruction feasibilityImproved
use event frame representation instead of voxel stack (ablation)class-wise noise · N-Caltech101 (ResNet50)test accuracy (%): 5.17No clear effect
use time surface representation instead of voxel stack (ablation)class-wise noise · N-Caltech101 (ResNet50)test accuracy (%): 10.09No clear effect
use C&W instead of PGD for noise (ablation)sample-wise noise · N-Caltech101 (ResNet50)test accuracy (%): 12.15No clear effect
replace UEvs with event error-minimizing noise vs polarity inversionUEvs sample-wise noise · N-Caltech101 (unlearnable test accuracy)test accuracy (%): 40.78 -> 0.52Improved

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