AI Paper Atlas

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

PureProof: Diffusion-Resistant Black-box Targeted Attack on Large Vision-Language Models

Yiming Cao, Dong Wang, Xinqi Lyu, Bin Xiao

CVPR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness

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 full PureProof (SRA+ARA+SCR) vs diffusion-unaware VLM attackstargeted VLM attacks · DiffPure defense, LLaVA-1.5ASR (Fool) %: 41.4 -> 76.8 (vs DH-cos)Improved
use single-step reverse alignment instead of full-chain backpropagation through diffusionPureProof · time per stepseconds per step: 44.717 -> 1.761Improved
add Adaptive Re-noising Augmentation (ARA) (ablation)PureProof · Qwen3-VL and LLaVA-1.5 vs DiffPureattack efficacyImproved
add SRA and SCR components (ablation)PureProof · Qwen3-VL and LLaVA-1.5 vs DiffPureattack efficacyImproved
increase re-noised variants K from 0 to 1 (ablation)PureProof ARA · Qwen3-VL and LLaVA-1.5 vs DiffPureattack efficacyImproved
increase K beyond 2 (ablation)PureProof ARA · Qwen3-VL and LLaVA-1.5 vs DiffPureattack efficacyNo clear effect
increase Gaussian noise from 8/255 to 16/255PureProof · MiniGPT-4 Gaussian noiseEnsemble CLIP score: 0.6342 -> 0.6315No clear effect
increase Gaussian noise from 8/255 to 16/255CoA · MiniGPT-4 Gaussian noiseEnsemble CLIP score: 0.6095 -> 0.4969Hurt
combine PureProof with CoA objectivePureProof · no defense, GPT-5Ensemble CLIP score: 0.6245 -> 0.6551Improved
use full-chain backpropagation (DA-cos/DH-cos) instead of PureProofdiffusion-aware attacks · DiffPure, Qwen3-VLASR (Fool) %: 38.3 -> 77.5 in favor of PureProofHurt

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