Image Classification on ImageNet 1k (val)
91Top-1 AccuracyCoCa
Evaluation results
| Method | Links | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CoCaEnc. #Params=1.01B, Patch Size=18 × 18, Image size=576^2, Pre-training Data=JFT-3B and ALIGN2023.05 | 91 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SotA (Fine-tuned)shots/class=full2022.04 | 90.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Model SoupsEnc. #Params=1.84B, Patch Size=14 × 14, Image size=500^2, Pre-training Data=JFT-3B2023.05 | 90.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-GEnc. #Params=1.84B, Patch Size=14 × 14, Image size=518^2, Pre-training Data=JFT-3B2023.05 | 90.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RevCol-HEnc. #Params=2.16B, Image size=640^2, Pre-training Data=Privately collected 168M-image dataset2023.05 | 90 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ONE-PEACEEnc. #Params=1.52B, Patch Size=16 × 16, Image size=512^22023.05 | 89.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EVAEnc. #Params=1.01B, Patch Size=14 × 14, Image size=560^22023.05 | 89.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-L/16 (Distilled from ViT-22B)Teacher=ViT-22B, Checkpoint=JFT ckpt., Resolution=384, Distillation Procedure=Beyer et al., 2022b2023.02 | 89.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BEIT-3Enc. #Params=1.01B, Patch Size=14 × 14, Image size=336^22023.05 | 89.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ONE-PEACEEnc. #Params=1.52B, Patch Size=16 × 16, Image size=384^22023.05 | 89.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FD-SwinV2-GEnc. #Params=3.0B, Patch Size=16 × 16, Image size=336^22023.05 | 89.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternImageEnc. #Params=1.08B, Patch Size=16 × 16, Image size=640^22023.05 | 89.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-B/16 (Distilled from ViT-22B)Teacher=ViT-22B, Checkpoint=JFT ckpt., Resolution=384, Distillation Procedure=Beyer et al., 2022b2023.02 | 88.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-L/16Reference=Zhai et al., 2022a, Checkpoint=JFT ckpt., Resolution=3842023.02 | 88.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Noisy Student (EfficientNet-L2)Params=480M, Extra Training Data=300M unlabeled images2020.03 | 88.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-L/16Reference=Touvron et al., 2022, Checkpoint=INet21k ckpt., Resolution=3842023.02 | 87.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SWIN-Large + DiversePatchInput Size=384, Pre-training=ImageNet-22K2021.04 | 87.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SWIN-LargeInput Size=384, Pre-training=ImageNet-22K2021.04 | 87.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-L/16Reference=Dosovitskiy et al., 2021, Checkpoint=JFT ckpt., Resolution=3842023.02 | 87.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-B/16Reference=Touvron et al., 2022, Checkpoint=INet21k ckpt., Resolution=3842023.02 | 86.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-B/16Reference=Zhai et al., 2022a, Checkpoint=JFT ckpt., Resolution=3842023.02 | 86.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F6#FLOPs=377.28B, #Params=438.4M, TPUv3 Train=2774.1ms, Sharpness-Aware Minimization (SAM)=true2021.02 | 86.5 | — | — | — | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F6+SAMParameters=438M, FLOPs=377.3B, Training Resolution=448, Test Resolution=5762021.03 | 86.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-M48↑448YParameters=356M, FLOPs=329.6B, Training Resolution=224, Test Resolution=4482021.03 | 86.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixRes ResNeXt-101 WSLParams=829M, Extra Training Data=3.5B tagged images2020.03 | 86.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F5#FLOPs=289.76B, #Params=377.2M, TPUv3 Train=1958.0ms, Sharpness-Aware Minimization (SAM)=true2021.02 | 86.3 | — | — | — | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-M36↑448YParameters=271M, FLOPs=247.8B, Training Resolution=224, Test Resolution=4482021.03 | 86.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-M36↑384YParameters=271M, FLOPs=173.3B, Training Resolution=224, Test Resolution=3842021.03 | 86.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F5#FLOPs=289.76B, #Params=377.2M, TPUv3 Train=1398.5ms, GPU Train=2177.1ms2021.02 | 86 | — | — | — | 97.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F5Parameters=377M, FLOPs=289.8B, Training Resolution=416, Test Resolution=5442021.03 | 86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F4#FLOPs=215.24B, #Params=316.1M, TPUv3 Train=1033.3ms, GPU Train=1190.6ms2021.02 | 85.9 | — | — | — | 97.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F4Parameters=316M, FLOPs=215.3B, Training Resolution=384, Test Resolution=5122021.03 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoAugment with Knowledge Distillation (EfficientNet-B8)backbone=EfficientNet-B8, Params=88M, Extra Training Data=None, Resolution=8002020.03 | 85.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| KDforAA-B8#param=87M, image size=800x800, throughput (image/s)=25.22020.12 | 85.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EffNet-B8#FLOPs=62.50B, #Params=87.4M, Augmentation=MaxUp2021.02 | 85.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F3#FLOPs=114.76B, #Params=254.9M, TPUv3 Train=532.2ms, GPU Train=524.5ms2021.02 | 85.7 | — | — | — | 97.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Fix-EfficientNet-B8Parameters=87M, FLOPs=89.5B, Training Resolution=672, Test Resolution=8002021.03 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F3Parameters=255M, FLOPs=114.8B, Training Resolution=320, Test Resolution=4162021.03 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BASICMethod=Contrastive, shots/class=02022.04 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-L2Params=480M, Extra Training Data=None2020.03 | 85.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdvProp (EfficientNet-B8)Params=88M, Extra Training Data=None2020.03 | 85.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoAugment with Knowledge Distillation (EfficientNet-B7)backbone=EfficientNet-B7, Params=66M, Extra Training Data=None2020.03 | 85.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAFormer-B36Type=H, #Param=99M, MACs=23.2G, Input Resolution=224x2242024.02 | 85.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B8Params=88M, Extra Training Data=None2020.03 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-S36↑384YParameters=68M, FLOPs=48.0B, Training Resolution=224, Test Resolution=3842021.03 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-Large + DiversePatchInput Size=384, Pre-training=ImageNet-22K2021.04 | 85.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-B↑384⚗ / 1000 epochs#param=87M, image size=384x384, throughput (image/s)=85.8, distillation=true, epochs=10002020.12 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeIT#Params=87.0M, Training Epochs=10002021.02 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 AdvPropParameters=66M, FLOPs=37.0B, Training Resolution=600, Test Resolution=6002021.03 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-B↑384Y 1000 epochsParameters=87M, FLOPs=55.5B, Training Resolution=224, Test Resolution=3842021.03 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hb Former-B36Type=C, #Param=102M, MACs=23.8G, Input Resolution=224x2242024.02 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-LType=H, #Param=212M, MACs=43.9G, Input Resolution=224x2242024.02 | 85.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F2#FLOPs=62.59B, #Params=193.8M, TPUv3 Train=295.8ms, GPU Train=226.3ms2021.02 | 85.1 | — | — | — | 97.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F2Parameters=194M, FLOPs=62.6B, Training Resolution=256, Test Resolution=3522021.03 | 85.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-S48↑384Parameters=89M, FLOPs=63.8B, Training Resolution=224, Test Resolution=3842021.03 | 85.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-LargeInput Size=384, Pre-training=ImageNet-22K2021.04 | 85.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAFormer-M36Type=H, #Param=56M, MACs=13.2G, Input Resolution=224x2242024.02 | 85.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7Params=66M, Extra Training Data=None2020.03 | 85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CaiT-S36↑384Parameters=68M, FLOPs=48.0B, Training Resolution=224, Test Resolution=3842021.03 | 85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-BType=H, #Param=120M, MACs=23.4G, Input Resolution=224x2242024.02 | 85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternImage-BType=C, #Param=97M, MACs=18.0G, Input Resolution=224x2242024.02 | 84.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hpx Former-B36Type=C, #Param=111M, MACs=25.3G, Input Resolution=224x2242024.02 | 84.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Billion-scale semi-supervised learning pipelineArchitecture=ResNeXt-101 32x16, Supervision Mode=semi-weakly sup.2019.05 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNeXt-101 Billion-scaleParams=193M, Extra Training Data=3.5B tagged images2020.03 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LambdaNet-420#Params=124.8M, TPUv3 Train=593.9ms2021.02 | 84.8 | — | — | — | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvFormer-B36Type=C, #Param=100M, MACs=22.6G, Input Resolution=224x2242024.02 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 RA#param=66M, image size=600x600, throughput (image/s)=55.12020.12 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F1#FLOPs=35.54B, #Params=132.6M, TPUv3 Train=158.5ms, GPU Train=133.9ms2021.02 | 84.7 | — | — | — | 97.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BoTNet-128-T7#FLOPs=45.80B, #Params=75.1M, TPUv3 Train=804.5ms2021.02 | 84.7 | — | — | — | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EffNet-B7#FLOPs=37.00B, #Params=66.0M, TPUv3 Train=1397.0ms, GPU Train=1753.3ms2021.02 | 84.7 | — | — | — | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7 RAParameters=66M, FLOPs=37.0B, Training Resolution=600, Test Resolution=6002021.03 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NFNet-F1Parameters=133M, FLOPs=35.5B, Training Resolution=224, Test Resolution=3202021.03 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hpx / Conv / Hb / CAType=H, #Param=111M, MACs=17.3G, Input Resolution=224x2242024.02 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-B↑384⚗#param=87M, image size=384x384, throughput (image/s)=85.8, distillation=true2020.12 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LambdaNet-350#Params=105.8M, TPUv3 Train=471.4ms2021.02 | 84.5 | — | — | — | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-SType=H, #Param=69M, MACs=11.7G, Input Resolution=224x2242024.02 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAFormer-S36Type=H, #Param=39M, MACs=8.0G, Input Resolution=224x2242024.02 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvFormer-M36Type=C, #Param=57M, MACs=12.8G, Input Resolution=224x2242024.02 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-B⚗ ↑384im/sec=85.9, resolution=384, distillation=true2020.12 | 84.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Billion-scale semi-supervised learning pipelineArchitecture=ResNeXt-101 32x8, Supervision Mode=semi-weakly sup.2019.05 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPipeParams=557M, Extra Training Data=None2020.03 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7im/sec=55.12020.12 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7#param=66M, image size=600x600, throughput (image/s)=55.12020.12 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B7Parameters=66M, FLOPs=37.0B, Training Resolution=600, Test Resolution=6002021.03 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficietNet-B7#Parameters=66M, FLOPs=37G, Resolution=600^2, Model Type=CNN, Design Type=Auto2021.07 | 84.3 | — | — | — | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiFormer-BType=H, #Param=59M, MACs=9.8G, Input Resolution=224x2242024.02 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-LType=C, #Param=198M, MACs=34.4G, Input Resolution=224x2242024.02 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Weakly supervisedArchitecture=ResNeXt-101 32x162019.05 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-B⚗ / 1000 epochs#param=87M, image size=224x224, throughput (image/s)=290.9, distillation=true, epochs=10002020.12 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| data2vecProtocol=Fine-tuning, Pre-train Dataset=ImageNet, Pre-train Type=Self-supervised (SSP intra-view), Backbone=ViT-B/16, Fine-tuning schedule=100 epochs2022.11 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M3I Pre-trainingProtocol=Fine-tuning, Pre-train Dataset=ImageNet, Pre-train Type=M3I, Backbone=ViT-B/16, Fine-tuning schedule=100 epochs2022.11 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-B/16Reference=Dosovitskiy et al., 2021, Checkpoint=JFT ckpt., Resolution=3842023.02 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternImage-SType=C, #Param=50M, MACs=8.0G, Input Resolution=224x2242024.02 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CSWin-BType=A, #Param=78M, MACs=15.0G, Input Resolution=224x2242024.02 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Next-DiTParams=86M, Setting=300E+30E, Resolution=Flexible2024.06 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiameseIMEpochs=1600, Evaluation Protocol=Full fine-tuning2022.06 | 84.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SiameseIMProtocol=Fine-tuning, Pre-train Dataset=ImageNet, Pre-train Type=Self-supervised (SSP inter-view), Backbone=ViT-B/16, Fine-tuning schedule=100 epochs2022.11 | 84.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientNet-B6#param=43M, image size=528x528, throughput (image/s)=96.92020.12 | 84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BoTNet-77-T6#FLOPs=23.30B, #Params=53.9M, TPUv3 Train=578.1ms2021.02 | 84 | — | — | — | 96.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EffNet-B6#FLOPs=19.00B, #Params=43.0M, TPUv3 Train=775.7ms, GPU Train=868.2ms2021.02 | 84 | — | — | — | 96.8 | — | — | — | — | — | — | — | — | — | — | — | — | — |