List your company

Visual Question Answering on VQA Average (VQAv2, OK-VQA, GQA)

66.9Accuracy (Clean)UniTok-MLLM
22.814434.259745.70557.1503Feb 20, 2026Mar 9, 2026Mar 26, 2026Apr 12, 2026Apr 29, 2026May 16, 2026Jun 2, 2026

Evaluation results

MethodLinks
UniTok-MLLMTokenizer=clean2026.0266.9106.4
UniTok-MLLMTokenizer=AT 4/2552026.0262.545.235.2
UniTok-MLLMTokenizer=AT 8/2552026.0259.147.139.4
UniTok-MLLMTokenizer=AT 12/2552026.0257.54740
UniTok-MLLMTokenizer=AT 16/2552026.0254.645.638.6
CLIPAttack Type=Clean2026.0654.55——
AdvXLCLIP-HAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.0624.51——