AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

VLBiasBench: A Comprehensive Benchmark for Evaluating Bias in Large Vision-Language Model

Sibo Wang, Xiangkui Cao, Jie Zhang, Zheng Yuan, Shiguang Shan, Xilin Chen, Wen Gao

TPAMI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 51 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
increase LLaVA-1.5 size from 7b to 13bLLaVA-1.5 · VLBiasBench open-ended raceRange VADER: 0.065 -> 0.044Improved
increase InstructBlip-flan-t5 language model from xl to xxlInstructBlip · VLBiasBench open-ended raceRange VADER: 0.225 -> 0.014Improved
move from text-dominant (Base) to image-dominant (Scene) evidenceLLaVA-1.5-13b · VLBiasBench close-endedACC: 0.578 -> 0.375Hurt
move from text-dominant (Base) to image-dominant (Scene) evidenceGemini · VLBiasBench close-endedACC: 0.672 -> 0.876Improved
add a second compared group via text-induced questions (Base to Text)InstructBlip-flan-t5-xxl · VLBiasBench close-endedACC: 0.772 -> 0.675Hurt
add a second compared group via text-induced questions (Base to Text)GPT-4o · VLBiasBench close-endedACC: 0.882 -> 0.773Hurt
add a second compared group via text-induced questions (Scene to Scene Text)InstructBlip-flan-t5-xxl · VLBiasBench close-endedACC: 0.616 -> 0.433Hurt
use ambiguous instead of disambiguated contextsLLaVA-1.5-7b · VLBiasBench close-ended BaseACC: 0.763 -> 0.001Hurt
use ambiguous instead of disambiguated contextsGemini · VLBiasBench close-ended (All)ACC: 0.553 -> 0.988Improved
apply sentiment thresholds to filter neutral responses (ablation)Range VADER metric · VLBiasBench open-endedrank stabilityNo clear effect
reverse the close-ended questions to test selection bias (ablation)close-ended benchmark · VLBiasBench close-endedmodel rank: 7 -> 6No clear effect
score sentiment with GPT instead of VADER (ablation)Range VADER metric · VLBiasBench open-ended racePearson correlation of ranks: 0.978No clear effect

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