List your company

Visual Question Answering on VQA v2 ($L\_inf$ Perturbation Accuracy)

74.5Accuracy (Clean)No defense
43.61251.63159.6567.669Jun 4, 2025

Evaluation results

MethodLinks
No defenseVLM=LLaVA 1.5-7B2025.0674.52.90
CLIPureVLM=LLaVA 1.5-7B2025.0673.34.62.1
FARE2VLM=LLaVA 1.5-7B2025.0671.734.923
DiffCAPVLM=LLaVA 1.5-7B2025.0670.369.168.5
JPEG-DLVLM=LLaVA 1.5-7B2025.0668.860.845.8
FARE4VLM=LLaVA 1.5-7B2025.0668.340.730.5
DiffPureVLM=LLaVA 1.5-7B2025.0667.365.866
TeCoA2VLM=LLaVA 1.5-7B2025.0666.933.821.8
TeCoA4VLM=LLaVA 1.5-7B2025.0663.24131.7
No defenseVLM=OF-9B2025.0648.51.80
CLIPureVLM=OF-9B2025.0647.318.817.5
FARE2VLM=OF-9B2025.06472417.2
JPEG-DLVLM=OF-9B2025.0646.839.532.4
DiffCAPVLM=OF-9B2025.0646.345.445.3
TeCoA2VLM=OF-9B2025.0646.223.520.5
FARE4VLM=OF-9B2025.0646.123.621
DiffPureVLM=OF-9B2025.0645.143.543.6
TeCoA4VLM=OF-9B2025.0644.823.621.3