AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

How Many Unicorns Are in This Image? A Safety Evaluation Benchmark for Vision LLMs

Haoqin Tu, Chenhang Cui, Zijun Wang, Yiyang Zhou, Bingchen Zhao, Junlin Han, Wangchunshu Zhou, Huaxiu Yao, Cihang Xie

ECCV 2024 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 110 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
append counterfactual suffix to OOD questions21 VLLMs · OODCV-CounterfactualOverall accuracy: average drop of 17.1% overall, 33.2% on Yes/NoHurt
counterfactual descriptions on GPT-4VGPT-4V · OODCV-CounterfactualOverall accuracy: 80.61 -> 69.00Hurt
use ensemble-free single CLIP ViT attack Mix.Attack instead of AttackBard at eps=64/255misleading attack on VLLMs · NIPS17 misleading attackMissing rate: +8.4% average over AttackBardImproved
use a mix of target text objects (Mix.Attack) instead of a single one (Sin.Attack) (ablation)CLIP ViT attack · NIPS17 misleading attackMissing rate: 79.7% -> 87.5% (LLaVA Vicuna-v0-7B)Improved
use less common sketch object categoriesVLLMs · Sketchy-VQA vs Sketchy-ChallengingF1: 4.4% average F1 dropHurt
apply direct vision jailbreak adversarial images to the three source VLLMsMiniGPT4, LLaVA, InstructBLIP · toxic output generationToxic output likelihood: 2.1x higher vs clean images; only ~5% marginal across all VLLMsImproved
vision-language training of LLM into VLLMLLaVA family · text jailbreak ASRAttack success rate: vanilla 1.3% -> 6.8%; white-box 77.5% -> 94.8%Hurt
stronger LLM in models with few tunable parametersMiniGPT4, InstructBLIP · OOD tasksOOD accuracy: average decrease 2.1% and 0.2%No clear effect
stronger LLM in fully fine-tuned modelsLLaVA, mPLUG-Owl · OOD tasksOOD accuracy: average boost 52.3% and 10.0%Improved
no explicit vision encoder (Fuyu)Fuyu · adversarial attack ASRASR: 33.5% lower ASR than other VLLMsImproved

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