List your company
ResearchFormerly WizWand

Image Captioning on MS COCO VIS

85.68ScoreClean
59.929666.614873.379.9852May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CleanImage Encoder=OpenAI CLIP ViT-L/14, Model=Blip-2 FlanT5XL ViT-L (3.4B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0585.68—
Unified PatchImage Encoder=OpenAI CLIP ViT-L/14, Model=Blip-2 FlanT5XL ViT-L (3.4B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0583.951.73
CleanImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.6 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0582.37—
Unified PatchImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.6 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0581.121.25
CleanImage Encoder=OpenAI CLIP ViT-L/14, Model=OpenFlamingo (3B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0579.46—
TOUAPImage Encoder=OpenAI CLIP ViT-L/14, Model=Blip-2 FlanT5XL ViT-L (3.4B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0579.46.28
CleanImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.5 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0578.53—
Unified PatchImage Encoder=OpenAI CLIP ViT-L/14, Model=OpenFlamingo (3B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0577.941.52
CleanImage Encoder=EVA-CLIP ViT-G/14, Model=InstructBLIP FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0577.56—
Unified PatchImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.5 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.05762.53
CFGPatchImage Encoder=OpenAI CLIP ViT-L/14, Model=Blip-2 FlanT5XL ViT-L (3.4B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0575.6510.03
Unified PatchImage Encoder=EVA-CLIP ViT-G/14, Model=InstructBLIP FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0575.641.92
CleanImage Encoder=EVA-CLIP ViT-G/14, Model=Blip-2 FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0575.38—
TOUAPImage Encoder=EVA-CLIP ViT-G/14, Model=InstructBLIP FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0572.115.45
TOUAPImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.5 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0571.956.58
Unified PatchImage Encoder=EVA-CLIP ViT-G/14, Model=Blip-2 FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0571.893.49
TOUAPImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.6 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0571.2511.12
TOUAPImage Encoder=OpenAI CLIP ViT-L/14, Model=OpenFlamingo (3B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0571.28.26
CFGPatchImage Encoder=EVA-CLIP ViT-G/14, Model=InstructBLIP FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0569.817.75
CFGPatchImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.6 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0567.215.17
CFGPatchImage Encoder=OpenAI CLIP ViT-L/14, Model=LLaVA-1.5 (7B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0565.3913.14
CFGPatchImage Encoder=OpenAI CLIP ViT-L/14, Model=OpenFlamingo (3B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0564.5314.93
TOUAPImage Encoder=EVA-CLIP ViT-G/14, Model=Blip-2 FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0563.8411.54
CFGPatchImage Encoder=EVA-CLIP ViT-G/14, Model=Blip-2 FlanT5XL (4.1B), Evaluation Protocol=LLM-as-Judge (GPT-5)2026.0560.9214.46