Image Classification on ImageNet ILSVRC-2012 (val)
90.88Top-1 AccuracyCoAtNet-7
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CoAtNet-7Params=2440M, Data=3000M, Fine-tuning=true2021.11 | 90.88 | — | — | — | |
| ViT-G/14Params=1843M, Data=3000M, Fine-tuning=true2021.11 | 90.45 | — | — | — | |
| Meta Pseudo LabelsBackbone=EfficientNet-L22020.03 | 90.2 | — | — | — | |
| Florence-CoSwin-HParams=637M, Data=900M, Fine-tuning=true2021.11 | 90.05 | 99.02 | — | — | |
| ALIGN-Efficient-L2Params=480M, Data=1800M, Fine-tuning=true2021.11 | 88.64 | 98.67 | — | — | |
| Previous SOTABackbone=EfficientNet-L22020.03 | 88.6 | — | — | — | |
| Noisy Student Training (EfficientNet-L2)# Params=480M, Extra Data=300M unlabeled images from JFT2019.11 | 88.4 | 98.7 | — | — | |
| BiT-L-ResNet152x4Params=928M, Data=300M, Fine-tuning=true2021.11 | 87.54 | 98.46 | — | — | |
| Big Transfer (BiT-L)# Params=928M, Extra Data=300M weakly labeled images from JFT2019.11 | 87.5 | 98.5 | — | — | |
| FixRes ResNeXt-101 WSL# Params=829M, Extra Data=3.5B images labeled with tags2019.11 | 86.4 | 98 | — | — | |
| EfficientNet-L2# Params=480M, Extra Data=-2019.11 | 85.5 | 97.5 | — | — | |
| ResNeXt-101 WSL# Params=829M, Extra Data=3.5B images labeled with tags2019.11 | 85.4 | 97.6 | — | — | |
| EfficientNet-B7# Params=66M, Extra Data=-2019.11 | 85 | 97.2 | — | — | |
| ResNeXt-101 Billion-scale# Params=193M, Extra Data=3.5B images labeled with tags2019.11 | 84.8 | — | — | — | |
| SSBackbone=Swin-B, Sampler Search Model=Target Model (Self)2024.10 | 84.3 | 96.7 | — | — | |
| GPipe# Params=557M, Extra Data=-2019.11 | 84.3 | 97 | — | — | |
| BaselineBackbone=Swin-B2024.10 | 83.5 | 96.5 | — | — | |
| AmoebaNet-C# Params=155M, Extra Data=-2019.11 | 83.5 | 96.5 | — | — | |
| Meta Pseudo LabelsUnlabeled Images=JFT, Backbone=ResNet-502020.03 | 83.2 | 96.5 | — | — | |
| EfficientNet-B4Resolution=380x380, Params (M)=19, FLOPs (B)=4.22021.02 | 82.9 | 96.4 | — | — | |
| TNT-BResolution=224x224, Params (M)=65.6, FLOPs (B)=14.12021.02 | 82.9 | 96.3 | — | — | |
| PNASNet# Params=86M, Extra Data=-2019.11 | 82.9 | 96.2 | — | — | |
| AmoebaNet-A# Params=87M, Extra Data=-2019.11 | 82.8 | 96.1 | — | — | |
| SENet# Params=146M, Extra Data=-2019.11 | 82.7 | 96.2 | — | — | |
| NASNet-A# Params=89M, Extra Data=-2019.11 | 82.7 | 96.2 | — | — | |
| Billion-scale SSLUnlabeled Images=YFCC, Backbone=ResNet-502020.03 | 82.5 | 96.6 | — | — | |
| ViG-BResolution=224x224, Architecture=GNN, Params (M)=86.8, FLOPs (B)=17.72022.06 | 82.3 | 95.9 | — | — | |
| T2T-ViT_t-24Resolution=224x224, Params (M)=63.9, FLOPs (B)=13.22021.02 | 82.2 | — | — | — | |
| SSBackbone=Swin-T, Sampler Search Model=Target Model (Self)2024.10 | 82.1 | 95.6 | — | — | |
| DeiT-BResolution=224x224, Params (M)=86.4, FLOPs (B)=17.62021.02 | 81.8 | — | — | — | |
| DeiT-BResolution=224x224, Architecture=Transformer, Params (M)=86.4, FLOPs (B)=17.62022.06 | 81.8 | 95.7 | — | — | |
| EfficientNet-B3Resolution=300x300, Params (M)=12, FLOPs (B)=1.82021.02 | 81.6 | 94.9 | — | — | |
| TNT-SResolution=224x224, Params (M)=23.8, FLOPs (B)=5.22021.02 | 81.5 | 95.7 | — | — | |
| ConvMixer-1536/20Resolution=224x224, Architecture=CNN, Params (M)=51.6, FLOPs (B)=51.42022.06 | 81.4 | — | — | — | |
| PolyNet# Params=92M, Extra Data=-2019.11 | 81.3 | 95.8 | — | — | |
| BaselineBackbone=Swin-T2024.10 | 81.2 | 95.5 | — | — | |
| ResNet-50 Billion-scale# Params=26M, Extra Data=3.5B images labeled with tags2019.11 | 81.2 | 96 | — | — | |
| ResMLP-B24Resolution=224x224, Architecture=MLP, Params (M)=116, FLOPs (B)=23.02022.06 | 81 | 95 | — | — | |
| ResNeXt-101# Params=84M, Extra Data=-2019.11 | 80.9 | 95.6 | — | — | |
| T2T-ViT_t-14Resolution=224x224, Params (M)=21.5, FLOPs (B)=5.22021.02 | 80.7 | — | — | — | |
| RegNetY-16GFResolution=224x224, Params (M)=83.6, FLOPs (B)=15.92021.02 | 80.4 | — | — | — | |
| ViG-SResolution=224x224, Architecture=GNN, Params (M)=22.7, FLOPs (B)=4.52022.06 | 80.4 | 95.2 | — | — | |
| PGT_FArch=PGT-B, Param.=115, Group Tokens=1024, Initialization=Flow, Protocol=Linear probe2023.11 | 80.3 | — | — | — | |
| ConvMixer-768/32Resolution=224x224, Architecture=CNN, Params (M)=21.1, FLOPs (B)=20.92022.06 | 80.2 | — | — | — | |
| WideNet-HParameters=63M2021.07 | 80.1 | — | — | — | |
| BEiTv2Arch=ViT-B/16, Param.=85, Protocol=Linear probe2023.11 | 80.1 | — | — | — | |
| DINOArch=ViT-B/8, Param.=85, Protocol=Linear probe2023.11 | 80.1 | — | — | — | |
| PGT_GArch=PGT-B, Param.=70, Group Tokens=1024, Initialization=Gaussian, Protocol=Linear probe2023.11 | 80.1 | — | — | — | |
| PGT_FArch=PGT-B, Param.=115, Group Tokens=512, Initialization=Flow, Protocol=Linear probe2023.11 | 80.1 | — | — | — | |
| Inception-resnet-v2# Params=56M, Extra Data=-2019.11 | 80.1 | 95.1 | — | — | |
| PGT_FArch=PGT-B, Param.=115, Group Tokens=256, Initialization=Flow, Protocol=Linear probe2023.11 | 80 | — | — | — | |
| SSBackbone=SE-ResNext-101, Sampler Search Model=Target Model (Self)2024.10 | 80 | 94.6 | — | — | |
| Inception-v4# Params=48M, Extra Data=-2019.11 | 80 | 95 | — | — | |
| RegNetY-8GFResolution=224x224, Params (M)=39.2, FLOPs (B)=82021.02 | 79.9 | — | — | — | |
| PGT_GArch=PGT-B, Param.=70, Group Tokens=512, Initialization=Gaussian, Protocol=Linear probe2023.11 | 79.9 | — | — | — | |
| FixRes+CutMixUnlabeled Images=None, Backbone=ResNet-502020.03 | 79.8 | 94.9 | — | — | |
| DeiT-SResolution=224x224, Params (M)=22.1, FLOPs (B)=4.62021.02 | 79.8 | — | — | — | |
| PVT-SmallResolution=224x224, Params (M)=24.5, FLOPs (B)=3.82021.02 | 79.8 | — | — | — | |
| DeiT-SResolution=224x224, Architecture=Transformer, Params (M)=22.1, FLOPs (B)=4.62022.06 | 79.8 | 95 | — | — | |
| SCLRv2Arch=RN152w3+SK, Param.=794, Protocol=Linear probe2023.11 | 79.8 | — | — | — | |
| PGT_GArch=PGT-S, Param.=34, Group Tokens=1024, Initialization=Gaussian, Protocol=Linear probe2023.11 | 79.8 | — | — | — | |
| SSBackbone=SE-ResNext-50, Sampler Search Model=Target Model (Self)2024.10 | 79.8 | 94.2 | — | — | |
| SSBackbone=SE-ResNext-101, Sampler Search Model=ResNet-182024.10 | 79.8 | 94.5 | — | — | |
| SSBackbone=SE-ResNext-101, Sampler Search Model=ResNet-502024.10 | 79.8 | 94.5 | — | — | |
| DINOArch=ViT-S/8, Param.=21, Protocol=Linear probe2023.11 | 79.7 | — | — | — | |
| PGT_GArch=PGT-B, Param.=70, Group Tokens=256, Initialization=Gaussian, Protocol=Linear probe2023.11 | 79.7 | — | — | — | |
| BYOLArch=RN200W2, Param.=250, Protocol=Linear probe2023.11 | 79.6 | — | — | — | |
| SSBackbone=SE-ResNext-50, Sampler Search Model=ResNet-502024.10 | 79.6 | 94.2 | — | — | |
| WideNet-LParameters=40M2021.07 | 79.5 | — | — | — | |
| iBOTArch=ViT-B/16, Param.=85, Protocol=Linear probe2023.11 | 79.5 | — | — | — | |
| SSBackbone=SE-ResNext-50, Sampler Search Model=ResNet-182024.10 | 79.5 | 94.2 | — | — | |
| ResMLP-S24Resolution=224x224, Architecture=MLP, Params (M)=30, FLOPs (B)=6.02022.06 | 79.4 | 94.5 | — | — | |
| SSBackbone=ResNet-101, Sampler Search Model=Target Model (Self)2024.10 | 79.3 | 94 | — | — | |
| ResNet-101 (Fine-tune from JFT-300M)Initialization=JFT-300M, Backbone=ResNet-101, Inference type=single model, single crop2017.07 | 79.2 | 94.7 | — | — | |
| FixResUnlabeled Images=None, Backbone=ResNet-502020.03 | 79.1 | 94.6 | — | — | |
| Xceptioncrop=single, model=single2016.10 | 79 | 94.5 | — | — | |
| SpineNet-143#FLOPs=9.1B, #Params=60.5M2019.12 | 79 | 94.4 | — | — | |
| UDAUnlabeled Images=JFT, Backbone=ResNet-502020.03 | 79 | 94.5 | — | — | |
| Xception# Params=23M, Extra Data=-2019.11 | 79 | 94.5 | — | — | |
| MaxUp+CutMixBackbone=ResNet-50, Evaluation setting=single-crop2020.02 | 78.9 | 94.2 | — | — | |
| Noisy StudentUnlabeled Images=JFT, Backbone=ResNet-502020.03 | 78.9 | 94.3 | — | — | |
| BaselineBackbone=SE-ResNext-1012024.10 | 78.9 | 94.4 | — | — | |
| Inception-v3# Params=24M, Extra Data=-2019.11 | 78.8 | 94.4 | — | — | |
| ResNet-152#FLOPs=11.5B, #Params=60.2M2019.12 | 78.7 | 94.2 | — | — | |
| CutMixBackbone=ResNet-50, Evaluation setting=single-crop2020.02 | 78.6 | 94 | — | — | |
| ViT-BParameters=87M2021.07 | 78.6 | — | — | — | |
| BYOLArch=RN50W4, Param.=375, Protocol=Linear probe2023.11 | 78.6 | — | — | — | |
| BaselineBackbone=SE-ResNext-502024.10 | 78.6 | 93.9 | — | — | |
| C2DArchitecture=ResNet-50, base_method=DivideMix with SimCLR2021.03 | 78.57 | 93.04 | — | — | |
| SwAVArch=RN50W5, Param.=586, Protocol=Linear probe2023.11 | 78.5 | — | — | — | |
| DropBlockUnlabeled Images=None, Backbone=ResNet-502020.03 | 78.4 | 94.2 | — | — | |
| DropBlock + label smoothingBackbone=ResNet-50, keep_prob=0.9, block_size=7, smoothing_factor=0.1, Training epochs=2702018.10 | 78.35 | 94.15 | — | — | |
| DropBlockBackbone=ResNet-50, Evaluation setting=single-crop2020.02 | 78.3 | 94.1 | — | — | |
| ResNet-152Resolution=224x224, Params (M)=60.2, FLOPs (B)=11.52021.02 | 78.3 | 94.1 | — | — | |
| Inception V3crop=single, model=single, auxiliary tower=false2016.10 | 78.2 | 94.1 | — | — | |
| SpineNet-96#FLOPs=5.7B, #Params=36.5M2019.12 | 78.2 | 94 | — | — | |
| ResNet-101#FLOPs=7.8B, #Params=44.6M2019.12 | 78.2 | 94.2 | — | — | |
| DINOArch=ViT-B/16, Param.=85, Protocol=Linear probe2023.11 | 78.2 | — | — | — | |
| DropBlockBackbone=ResNet-50, keep_prob=0.9, block_size=7, Training epochs=2702018.10 | 78.13 | 94.02 | — | — | |
| BaselineBackbone=ResNet-1012024.10 | 78 | 93.8 | — | — |