AI Paper Atlas

Research map / Vision-language and multimodal / Visual question answering

CBV: Clean-label Backdoor Attacks on Vision Language Models via Diffusion Models

Ji Guo, xiaolong qin, Cencen Liu, Jielei Wang, Jierun Chen, Wenbo Jiang

ICML 2026 · Vision-language and multimodal › Visual question answering · approach: Vision-language model VQA evaluation · 0 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 text guidance to diffusion score modification (ablation)CBV · VQA v2, LLaVA-v1.5-7BASR (%): 74.47 -> 85.77Improved
add text guidance to diffusion score modification (ablation)CBV · VQA v2, MiniGPT-v2ASR (%): 72.09 -> 83.83Improved
use UAP trigger instead of patch (ablation)CBV · MSCOCOASR (%): 78.40 -> 84.27Improved
use UAP trigger instead of noise (ablation)CBV · MSCOCOASR (%): 74.16 -> 84.27Improved
add GradCAM-guided mask (ablation)CBV · MSCOCOSSIM: 0.5511 -> 0.7139Improved
add GradCAM-guided mask (ablation)CBV · MSCOCOLPIPS: 26.52 -> 23.78Improved
use CLIP-based surrogates versus ResNet-50 for the GradCAM mask (ablation)CBV · MSCOCO captioningASR (%)No clear effect
increase poisoning rate (ablation)CBV · LLaVA-v1.5-7BASR (%) vs clean performance: 60% ASR at 2%, near 80% at 4%Mixed
use diffusion-based generation instead of surrogate-optimized clean-label poisoningCBV vs CBAVR/Narcissus/HTBA/SAA/IP · MSCOCO captioning, LLaVA-v1.5-7BASR (%): Narcissus 11.54 -> 85.77Improved
use CBV instead of dirty-label attacksCBV · MSCOCO captioning, LLaVA-v1.5-7BASR (%): TrojVLM 94.60 vs CBV 85.77Mixed
apply fine-pruning defense up to 10% pruningCBV · backdoored VLMASRNo clear effect
increase text guidance weight lambda_T excessively (ablation)CBV · MSCOCO / VQA v2ASR and CIDErHurt

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