List your company
ResearchFormerly WizWand

Text-to-Image Retrieval on Cola (test)

83.88Multi-Obj AccHuman
18.921635.785852.6569.5142May 5, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Humantype=human performance2023.0583.88
MM-PredBase Model=CLIP, Adaptation Strategy=MM-Pred (multimodal head)2023.0541.42
MM-AdapterBase Model=CLIP, Adaptation Strategy=MM-Adapter (multimodal feature adapter)2023.0540.95
MM-AdapterBase Model=FLAVA, Adaptation Strategy=MM-Adapter (multimodal feature adapter)2023.0540.47
MM-PredBase Model=FLAVA, Adaptation Strategy=MM-Pred (multimodal head)2023.0539.04
CLIP + FT lateBase Model=CLIP, Adaptation Strategy=Late Fine-tuning2023.0536.19
CLIP + FT allBase Model=CLIP, Adaptation Strategy=Full Fine-tuning2023.0534.76
CLIP + LinearBase Model=CLIP, Adaptation Strategy=Linear2023.0530.47
CLIP + Prompt-tuneBase Model=CLIP, Adaptation Strategy=Prompt-tuning2023.0527.14
Randomtype=baseline2023.0525
FLAVABase Model=FLAVA2023.0524.76
FLAVA + LinearBase Model=FLAVA, Adaptation Strategy=Linear2023.0522.38
FLAVA + FT lateBase Model=FLAVA, Adaptation Strategy=Late Fine-tuning2023.0522.38
CLIPBase Model=CLIP2023.0521.42