List your company
ResearchFormerly WizWand

Image Classification on 27 image classification datasets average

80.7Average AccuracyEVA-CLIP-18B
71.8674.15576.4578.745Feb 6, 2024
Updated 8mo ago

Evaluation results

MethodLinks
EVA-CLIP-18Btotal #param.=18.1B, image #param.=17.5B, text #param.=695M, data=Merged-2B+, samples seen=6B, image size=224^2, batch size=108k, gpus for training=360×A100 (40GB), Zero-shot=true2024.0280.7
EVA-CLIP-18BZero-shot=true2024.0280.7
EVA-CLIP-8Btotal #param.=8.1B, image #param.=7.5B, text #param.=695M, data=Merged-2B [63], samples seen=9B, image size=224^2, batch size=178k, gpus for training=384×A100 (40GB), Zero-shot=true2024.0279.4
EVA-CLIP-8BZero-shot=true2024.0279.4
DFN5B-CLIP-H/14+total #param.=1.0B, image #param.=632M, text #param.=354M, data=DFN-5B [28], samples seen=5B, image size=378^2, batch size=79k, gpus for training=TPUv4, Zero-shot=true2024.0279.2
DFN5B-CLIP-H/14+Zero-shot=true2024.0279.2
EVA-02-CLIP-E/14+total #param.=5.0B, image #param.=4.4B, text #param.=695M, data=LAION-2B [58], samples seen=9B, image size=224^2, batch size=144k, gpus for training=144×A100 (80GB), Zero-shot=true2024.0278.7
EVA-02-CLIP-E/14+Zero-shot=true2024.0278.7
DFN5B-CLIP-H/14total #param.=1.0B, image #param.=632M, text #param.=354M, data=DFN-5B [28], samples seen=39B, image size=224^2, batch size=79k, gpus for training=TPUv4, Zero-shot=true2024.0278.3
DFN5B-CLIP-H/14Zero-shot=true2024.0278.3
InternVL-Ctotal #param.=14.0B, image #param.=6.0B, text #param.=8.0B, data=custom [17], samples seen=29B, image size=224^2, batch size=164k, gpus for training=640×A100 (80GB), Zero-shot=true2024.0278
InternVL-CZero-shot=true2024.0278
OpenCLIP-G/14total #param.=2.5B, image #param.=1.8B, text #param.=695M, data=LAION-2B [58], samples seen=39B, image size=224^2, batch size=160k, gpus for training=512×A100 (80GB), Zero-shot=true2024.0276.2
OpenCLIP-G/14Zero-shot=true2024.0276.2
EVA-01-CLIP-g/14+total #param.=1.3B, image #param.=1.0B, text #param.=354M, data=Merged-2B [63], samples seen=11B, image size=224^2, batch size=114k, gpus for training=112×A100 (40GB), Zero-shot=true2024.0275.1
EVA-01-CLIP-g/14+Zero-shot=true2024.0275.1
EVA-01-CLIP-g/14total #param.=1.1B, image #param.=1.0B, text #param.=124M, data=LAION-400M [59], samples seen=11B, image size=224^2, batch size=41k, gpus for training=256×A100 (40GB), Zero-shot=true2024.0272.2
EVA-01-CLIP-g/14Zero-shot=true2024.0272.2