List your company

Multimodal Compositional Understanding on ARO

86.7Relational ScoreCapPa
37.8250.5163.275.89May 23, 2023May 26, 2023May 30, 2023Jun 2, 2023Jun 6, 2023Jun 9, 2023Jun 13, 2023

Evaluation results

MethodLinks
CapPamode=Zero-shot2023.0686.785.7—99.298.8
Capmode=Zero-shot2023.0686.688.9—99.199
Blind dec.mode=Zero-shot2023.0686.283.7—98.898.7
MosaiCLIPArch.=Swin-T, Pre-training data=CC-12M2023.0584.376.855.5——
MosaiCLIP2023.0583.77879.4——
MosaiCLIP NoCurriccurriculum_learning=false2023.0583.278.677.3——
MosaiCLIP WiSE-FTfine_tuning=WiSE-FT2023.0582.67779.9——
NegCLIPArch.=Swin-T, Pre-training data=CC-12M2023.0582.466.859.7——
MosaiCLIPArch.=RN-50, Pre-training data=CC-12M2023.058278.555.4——
NegCLIPmode=Zero-shot2023.068171—9186
NegCLIPArch.=RN-50, Pre-training data=CC-12M2023.0580.566.560.5——
MosaiCLIPArch.=RN-50, Pre-training data=YFCC-15M2023.0576.368.938.2——
MosaiCLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.0574.766.135.8——
NegCLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.0573.658.935.5——
ARO Bestmode=Zero-shot2023.067388—6046
Syn-CLIPtraining_data=Million-scale synthetic data2023.0571.466.965.1——
NegCLIPArch.=RN-50, Pre-training data=YFCC-15M2023.056858.537.1——
CLIPmode=Zero-shot2023.0559.963.653.3——
CLIPmode=Zero-shot2023.0658.762.7—57.949.5
CLIPArch.=RN-50, Pre-training data=YFCC-15M2023.0557.855.118.3——
CLIPArch.=Swin-T, Pre-training data=YFCC-15M2023.0553.856.218.4——
CLIPArch.=RN-50, Pre-training data=CC-12M2023.0552.959.722.6——
CLIPArch.=Swin-T, Pre-training data=CC-12M2023.055156.625.5——
RandomPre-training data=CC-12M2023.05505020——
RandomPre-training data=YFCC-15M2023.05505020——
CLIP-FTmode=Fine-tuned2023.0541.759.325.2——
CLIP*mode=Zero-shot2023.0639.753.2—45.537