List your company
ResearchFormerly WizWand

Vision-Language In-Context Learning on VL-ICL Bench Fast Open MiniImageNet (test)

100Top-1 AccuracyReVisIT (GPT-4.1)
16.2838.01559.7581.485Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
ReVisIT (GPT-4.1)K (shots per class)=22026.07100
ReVisIT (GPT-4.1)K (shots per class)=42026.07100
ReVisIT (Doubao-1.5)K (shots per class)=42026.07100
ReVisIT (Doubao-1.5)K (shots per class)=22026.0799.5
LLaVA-OneVision-72BK (shots per class)=42026.0798.7
ReVisIT (Qwen3-VL-30B, open)K (shots per class)=42026.0798.5
ReVisIT (Doubao-1.5)K (shots per class)=12026.0797.5
ReVisIT (GPT-4.1)K (shots per class)=12026.0796.5
ReVisIT (Qwen3-VL-30B, open)K (shots per class)=22026.0795
ReVisIT (Qwen3-VL-30B, open)K (shots per class)=12026.0780
GPT-4VK (shots per class)=42026.0778
IDEFICS-80BK (shots per class)=42026.0762.5
Qwen-VL-ChatK (shots per class)=42026.0758
ReVisIT (Qwen3-VL-30B, open)K (shots per class)=02026.0719.5