AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Manifold Projection for Adversarial Defense on Face Recognition

Jianli Zhou, Chao Liang, Jun Chen

ECCV 2020 · Adversarial robustness and security › Attacks on text and graphs · approach: Biologically inspired robust vision · 33 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 A-VAE input purificationResNet-50 face verifier · LFW grey-box FGSM eps=8verification accuracy (average): 0.245 -> 0.753Improved
A-VAE instead of adversarial trainingResNet-50 face verifier · LFW grey-box FGSM eps=8verification accuracy (average): 0.457 -> 0.753Improved
add A-VAE input purificationResNet-50 face verifier · LFW grey-box PGD eps=8verification accuracy (average): 0.003 -> 0.828Improved
add A-VAE input purificationResNet-50 face verifier · LFW grey-box C&Wverification accuracy (average): 0.008 -> 0.610Improved
add A-VAE input purificationResNet-50 face verifier · LFW clean imagesverification accuracy (average): 0.992 -> 0.963Hurt
attacker knows defense (white-box via BPDA) instead of grey-boxA-VAE · LFW PGD eps=8verification accuracy (average): 0.828 -> 0.573Hurt
attacker knows defense (white-box via BPDA) instead of grey-boxA-VAE · LFW FGSM eps=8verification accuracy (average): 0.753 -> 0.552Hurt
add A-VAE input purificationResNet-50 face verifier · LFW on-manifold attackverification accuracy (average): 0.321 -> 0.667Improved
use A-VAE rather than HGDResNet-50 face verifier · LFW white-box PGD eps=8verification accuracy (average): 0.518 -> 0.573Improved
increase generator input size from 32x32 to 128x128 (ablation)A-VAE · LFW grey-box FGSM eps=8robustness vs reconstruction qualityMixed
replace pixel-wise reconstruction loss with adversarial loss (ablation)A-VAE · LFWrobustnessImproved
downsample input to 32x32 before encoder (ablation)A-VAE · LFWrobustness and fidelityImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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