Image Classification on ILSVRC 2012 (test)
88.4Top-1 AccSpecialist SOTA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Specialist SOTAPre-training strategy=Specialist2019.12 | 88.4 | — | — | — | |
| BiT-LPre-training strategy=Generalist, Hyperparameter setting=BiT-HyperRule, Evaluation protocol=Fine-tuning2019.12 | 87.54 | — | — | — | |
| EfficientNetV2-XL (21k)Params=208M, FLOPs=94B, Train-time (hours)=45, Pre-training=ImageNet21k2021.04 | 87.3 | — | — | — | |
| EfficientNetV2-L (21k)Params=120M, FLOPs=53B, Infer-time (ms)=98, Train-time (hours)=26, Pre-training=ImageNet21k2021.04 | 86.8 | — | — | — | |
| Generalist SOTAPre-training strategy=Generalist2019.12 | 86.4 | — | — | — | |
| EfficientNetV2-M (21k)Params=54M, FLOPs=24B, Infer-time (ms)=57, Train-time (hours)=15, Pre-training=ImageNet21k2021.04 | 86.2 | — | — | — | |
| NFNet-F4Params=316M, FLOPs=215B, Infer-time (ms)=309, Train-time (hours)=126, Pre-training=Scratch2021.04 | 85.9 | — | — | — | |
| NFNet-F3Params=255M, FLOPs=115B, Infer-time (ms)=203, Train-time (hours)=65, Pre-training=Scratch2021.04 | 85.7 | — | — | — | |
| EfficientNetV2-LParams=120M, FLOPs=53B, Infer-time (ms)=98, Train-time (hours)=24, Pre-training=Scratch2021.04 | 85.7 | — | — | — | |
| ViT-L/16 (21k)Params=304M, FLOPs=192B, Infer-time (ms)=195, Train-time (hours)=172, Pre-training=ImageNet21k2021.04 | 85.3 | — | — | — | |
| BiT-M-R152x2Params=236M, FLOPs=135B, Infer-time (ms)=500, Pre-training=ImageNet21k2021.04 | 85.2 | — | — | — | |
| NFNet-F2Params=194M, FLOPs=63B, Infer-time (ms)=124, Train-time (hours)=36, Pre-training=Scratch2021.04 | 85.1 | — | — | — | |
| EfficientNetV2-MParams=54M, FLOPs=24B, Infer-time (ms)=57, Train-time (hours)=13, Pre-training=Scratch2021.04 | 85.1 | — | — | — | |
| LambdaResNet-420-hybridParams=125M, Train-time (hours)=67, Pre-training=Scratch2021.04 | 84.9 | — | — | — | |
| EfficientNetV2-S (21k)Params=22M, FLOPs=8.8B, Infer-time (ms)=24, Train-time (hours)=9, Pre-training=ImageNet21k2021.04 | 84.9 | — | — | — | |
| EfficientNet-B7Params=66M, FLOPs=38B, Infer-time (ms)=170, Train-time (hours)=139, Pre-training=Scratch2021.04 | 84.7 | — | — | — | |
| EfficientNet-XParams=73M, FLOPs=91B, Pre-training=Scratch2021.04 | 84.7 | — | — | — | |
| NFNet-F1Params=133M, FLOPs=36B, Infer-time (ms)=70, Train-time (hours)=20, Pre-training=Scratch2021.04 | 84.7 | — | — | — | |
| BotNet-T7-hybridParams=75M, FLOPs=46B, Train-time (hours)=95, Pre-training=Scratch2021.04 | 84.7 | — | — | — | |
| ViT-B/16 (21k)Params=87M, FLOPs=56B, Infer-time (ms)=68, Pre-training=ImageNet21k2021.04 | 84.6 | — | — | — | |
| ResNeSt-269Params=111M, FLOPs=78B, Infer-time (ms)=160, Pre-training=Scratch2021.04 | 84.5 | — | — | — | |
| EfficientNet-B6Params=43M, FLOPs=19B, Infer-time (ms)=97, Train-time (hours)=75, Pre-training=Scratch2021.04 | 84.3 | — | — | — | |
| TResNet-XLParams=78M, Infer-time (ms)=66, Pre-training=Scratch2021.04 | 84.3 | — | — | — | |
| ResNeSt-200Params=70M, FLOPs=36B, Infer-time (ms)=76, Pre-training=Scratch2021.04 | 83.9 | — | — | — | |
| EfficientNetV2-SParams=22M, FLOPs=8.8B, Infer-time (ms)=24, Train-time (hours)=7.1, Pre-training=Scratch2021.04 | 83.9 | — | — | — | |
| TResNet-LParams=56M, Infer-time (ms)=45, Pre-training=Scratch2021.04 | 83.8 | — | — | — | |
| EfficientNet-B5Params=30M, FLOPs=10B, Infer-time (ms)=60, Train-time (hours)=43, Pre-training=Scratch2021.04 | 83.7 | — | — | — | |
| NFNet-F0Params=72M, FLOPs=12B, Infer-time (ms)=30, Train-time (hours)=8.9, Pre-training=Scratch2021.04 | 83.6 | — | — | — | |
| DeiT-B-384Params=86M, FLOPs=56B, Infer-time (ms)=68, Pre-training=Scratch2021.04 | 83.1 | — | — | — | |
| ResNeSt-101Params=48M, FLOPs=13B, Infer-time (ms)=31, Pre-training=Scratch2021.04 | 83 | — | — | — | |
| EfficientNet-B4Params=19M, FLOPs=4.2B, Infer-time (ms)=30, Train-time (hours)=21, Pre-training=Scratch2021.04 | 82.9 | — | — | — | |
| RegNetY-16GFParams=84M, FLOPs=16B, Infer-time (ms)=32, Pre-training=Scratch2021.04 | 82.9 | — | — | — | |
| T2T-ViT-24Params=64M, FLOPs=13B, Pre-training=Scratch2021.04 | 82.2 | — | — | — | |
| DeiT-BParams=86M, FLOPs=18B, Infer-time (ms)=19, Pre-training=Scratch2021.04 | 81.8 | — | — | — | |
| RegNetY-8GFParams=39M, FLOPs=8B, Infer-time (ms)=21, Pre-training=Scratch2021.04 | 81.7 | — | — | — | |
| EfficientNet-B3Params=12M, FLOPs=1.9B, Infer-time (ms)=19, Train-time (hours)=10, Pre-training=Scratch2021.04 | 81.5 | — | — | — | |
| T2T-ViT-19Params=39M, FLOPs=8.4B, Pre-training=Scratch2021.04 | 81.4 | — | — | — | |
| MOAMlabels=100%, status=proposed2019.05 | 80.17 | — | 94.97 | — | |
| Jo-SNCBackbone=RN502026.01 | 80.16 | — | 93.88 | — | |
| Jo-SNCBackbone=IRNv22026.01 | 80.08 | — | 93.84 | — | |
| Modified ModelModel Variant=Modified Model, Test-Time Augmentation=true2022.10 | 79.64 | — | 94.2 | — | |
| C2D-GMBackbone=RN502026.01 | 79.16 | — | 93.12 | — | |
| UDAlabels=100%2019.05 | 79.04 | — | 94.45 | — | |
| ResNet50v2labels=100%, width=4x wider2019.05 | 78.57 | — | 94.1 | — | |
| DivideMix+C2DBackbone=RN502026.01 | 78.57 | — | 93.04 | — | |
| Bootstrapping Training AlgorithmModel Variant=Normal Model, Test-Time Augmentation=true2022.10 | 78.36 | — | 91.8 | — | |
| Ours2024.08 | 77.8 | — | 92.08 | — | |
| DARTS-Multiply-Adds (M)=470, Parameters (M)=5.52019.08 | 77.8 | — | 93.9 | — | |
| FairNAS-AMultiply-Adds (M)=392, Parameters (M)=5.92019.08 | 77.5 | — | 93.7 | — | |
| CMW-Net-SL+C2DBackbone=RN502026.01 | 77.36 | — | 93.48 | — | |
| FairDARTS-CMultiply-Adds (M)=386, Parameters (M)=5.32019.08 | 77.2 | — | 93.5 | — | |
| FaMuS2022.10 | 77 | — | 92.76 | — | |
| FaMUS2024.08 | 77 | — | 92.8 | — | |
| SCARLET-AMultiply-Adds (M)=365, Parameters (M)=6.7, Trained from scratch without AutoAugment=true2019.08 | 76.9 | — | 93.4 | — | |
| SelCL+2024.08 | 76.8 | — | 93 | — | |
| MixNet-MMultiply-Adds (M)=360, Parameters (M)=5, Trained from scratch without AutoAugment=true2019.08 | 76.6 | — | 93.2 | — | |
| Bootstrapping Training AlgorithmModel Variant=Normal Model, Test-Time Augmentation=false2022.10 | 76.44 | — | 90 | — | |
| NCEBackbone=IRNv22026.01 | 76.3 | — | 94.1 | — | |
| EfficientNet B0Multiply-Adds (M)=390, Parameters (M)=5.3, Trained from scratch without AutoAugment=true2019.08 | 76.3 | — | 93.2 | — | |
| SCARLET-BMultiply-Adds (M)=329, Parameters (M)=6.5, Trained from scratch without AutoAugment=true2019.08 | 76.3 | — | 93 | — | |
| DivideMix-GMBackbone=RN502026.01 | 76.11 | — | 93.24 | — | |
| Modified ModelModel Variant=Modified Model, Test-Time Augmentation=false2022.10 | 75.96 | — | 92.2 | — | |
| SSR+2024.08 | 75.8 | — | 91.8 | — | |
| ScanMixEpochs=1002021.03 | 75.76 | — | 92.6 | — | |
| CMW-Net-SLBackbone=RN502026.01 | 75.72 | — | 92.52 | — | |
| MnasNet-A2Multiply-Adds (M)=340, Parameters (M)=4.82019.08 | 75.6 | — | 92.7 | — | |
| SCARLET-CMultiply-Adds (M)=280, Parameters (M)=6, Trained from scratch without AutoAugment=true2019.08 | 75.6 | — | 92.6 | — | |
| TCL2024.08 | 75.4 | — | 92.4 | — | |
| UNICONBackbone=IRNv22026.01 | 75.29 | — | 93.72 | — | |
| DivideMixEpochs=1002021.03 | 75.2 | — | 90.84 | — | |
| DivideMix2024.08 | 75.2 | — | 90.84 | — | |
| DivideMixBackbone=IRNv22026.01 | 75.2 | — | 90.84 | — | |
| MobileNetV3Multiply-Adds (M)=219, Parameters (M)=5.42019.08 | 75.2 | — | 92.2 | — | |
| MnasNet-A1Multiply-Adds (M)=312, Parameters (M)=3.92019.08 | 75.2 | — | 92.5 | — | |
| Proxyless GPUMultiply-Adds (M)=465, Parameters (M)=7.12019.08 | 75.1 | — | — | — | |
| Single-PathMultiply-Adds (M)=365, Parameters (M)=4.32019.08 | 75 | — | 92.2 | — | |
| ViT-B/16Params=87M, FLOPs=56B, Infer-time (ms)=68, Pre-training=Scratch2021.04 | 74.9 | — | — | — | |
| SPOSMultiply-Adds (M)=328, Parameters (M)=3.42019.08 | 74.9 | — | 92 | — | |
| Proxyless-RMultiply-Adds (M)=320, Parameters (M)=42019.08 | 74.6 | — | 92.2 | — | |
| NGC2022.10 | 74.44 | — | 91.04 | — | |
| NGC2024.08 | 74.44 | — | 91.04 | — | |
| DivideMixBackbone=RN502026.01 | 74.42 | — | 91.21 | — | |
| FBNet-BMultiply-Adds (M)=295, Parameters (M)=4.52019.08 | 74.1 | — | — | — | |
| ViT-B/32Params=88M, FLOPs=13B, Infer-time (ms)=13, Pre-training=Scratch2021.04 | 73.4 | — | — | — | |
| RRL2022.10 | 73.3 | — | 91.2 | — | |
| RRL2024.08 | 73.3 | — | 91.2 | — | |
| MOAM (full)labels=10%, status=proposed2019.05 | 73.21 | — | 91.23 | — | |
| MentorMixEpochs=1002021.03 | 72.9 | — | 91.1 | — | |
| FSREpochs=1002021.03 | 72.3 | — | 87.2 | — | |
| MobileNetV2Multiply-Adds (M)=300, Parameters (M)=3.42019.08 | 72 | — | 91 | — | |
| MOAM (+ pseudo label)labels=10%, status=proposed2019.05 | 71.56 | — | 89.96 | — | |
| ELR+Epochs=1002021.03 | 70.29 | — | 89.76 | — | |
| ELR+2022.10 | 70.29 | — | 89.76 | — | |
| ELR+2024.08 | 70.29 | — | 89.76 | — | |
| ELR+Backbone=IRNv22026.01 | 70.29 | — | 89.76 | — | |
| MOAMlabels=10%, status=proposed2019.05 | 69.73 | — | 88.8 | — | |
| SOPBackbone=IRNv22026.01 | 69.1 | — | — | — | |
| ELR2022.10 | 68.71 | — | 87.84 | — | |
| ELRBackbone=IRNv22026.01 | 68.71 | — | 87.84 | — | |
| UDAlabels=10%2019.05 | 68.66 | — | 88.52 | — |