List your company

Image Captioning on COCO (test) (L-infinity Metrics)

120.4Clean Metric ValueDiffCAP
64.7679.20593.65108.095Jun 4, 2025

Evaluation results

MethodLinks
DiffCAPBackbone=LLaVA 1.5-7B2025.06120.4119.6116.9
No defenseBackbone=LLaVA 1.5-7B2025.06115.543.1
CLIPureBackbone=LLaVA 1.5-7B2025.06115.14.93.4
JPEG-DLBackbone=LLaVA 1.5-7B2025.06113.3106.477.2
FARE2Backbone=LLaVA 1.5-7B, fine-tuning perturbation=ℓ∞_2/2552025.06109.953.631
DiffPureBackbone=LLaVA 1.5-7B2025.06106.5108.4105
FARE4Backbone=LLaVA 1.5-7B, fine-tuning perturbation=ℓ∞_4/2552025.06102.457.140.9
TeCoA2Backbone=LLaVA 1.5-7B, fine-tuning perturbation=ℓ∞_2/2552025.0698.444.230.3
TeCoA4Backbone=LLaVA 1.5-7B, fine-tuning perturbation=ℓ∞_4/2552025.0688.350.935.3
DiffCAPBackbone=OF-9B2025.0681.479.378.4
CLIPureBackbone=OF-9B2025.0680.86.65.3
No defenseBackbone=OF-9B2025.0679.71.51.1
FARE2Backbone=OF-9B, fine-tuning perturbation=ℓ∞_2/2552025.0679.134.219.5
JPEG-DLBackbone=OF-9B2025.0678.266.143.9
DiffPureBackbone=OF-9B2025.0674.973.472.3
FARE4Backbone=OF-9B, fine-tuning perturbation=ℓ∞_4/2552025.0674.130.922.8
TeCoA2Backbone=OF-9B, fine-tuning perturbation=ℓ∞_2/2552025.0673.531.621.2
TeCoA4Backbone=OF-9B, fine-tuning perturbation=ℓ∞_4/2552025.0666.928.521.6