Image Classification on ILSVRC 2012 (val)
84.5Top-1 AccuracyEfficientNet B7
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| EfficientNet B7Throughput=16, Augmentation=AutoAugment2020.01 | 84.5 | — | — | — | — | 59.4 | — | — | — | — | — | |
| EfficientNet B6Throughput=28, Augmentation=AutoAugment2020.01 | 84.2 | — | — | — | — | 60.6 | — | — | — | — | — | |
| Assemble-ResNet-152Throughput=1432020.01 | 84.2 | — | — | — | — | 43.3 | 29.3 | — | — | — | — | |
| EfficientNet B4Throughput=95, Augmentation=AutoAugment2020.01 | 83 | — | — | — | — | 60.7 | — | — | — | — | — | |
| Assemble-ResNet-50Throughput=3122020.01 | 82.8 | — | — | — | — | 48.9 | 32.3 | — | — | — | — | |
| Heteroscedastic FullBackbone=ResNet-152, Epochs=270, Temperature (tau*)=0.92021.05 | 79.3 | — | — | — | 94.5 | — | — | 0.92 | — | — | — | |
| Het. DiagBackbone=ResNet-152, Epochs=270, Temperature (tau*)=0.92021.05 | 78.7 | — | — | — | 94 | — | — | 0.95 | — | — | — | |
| DivideMix with C2DBackbone=ResNet-50, Pretraining=C2D2022.02 | 78.57 | — | — | — | 93.04 | — | — | — | — | — | — | |
| Heteroscedastic FullBackbone=ResNet-152, Epochs=90, Temperature (tau*)=0.92021.05 | 78.5 | — | — | — | 94.3 | — | — | 0.86 | — | — | — | |
| Resnet-200Params=64.7M, Evaluation protocol=single crop2017.03 | 78.3 | — | — | — | 94.2 | — | — | — | — | — | — | |
| Het. DiagBackbone=ResNet-152, Epochs=90, Temperature (tau*)=0.92021.05 | 78.3 | — | — | — | 94 | — | — | 0.88 | — | — | — | |
| CMW-Net-SL+C2DBackbone=ResNet-50, Pretraining=C2D2022.02 | 77.36 | — | — | — | 93.48 | — | — | — | — | — | — | |
| CoDiM-Self2021.11 | 77.24 | — | — | — | 92.48 | — | — | — | — | — | — | |
| CoDiMvariant=self2021.11 | 77.24 | — | — | — | 92.48 | — | — | — | — | — | — | |
| Sel-CL+2022.03 | 76.84 | — | — | — | 93.04 | — | — | — | — | — | — | |
| HomoscedasticBackbone=ResNet-152, Epochs=2702021.05 | 76.7 | — | — | — | 92.9 | — | — | 1.08 | — | — | — | |
| C2Dimplementation=re-implemented2021.11 | 76.64 | — | — | — | 92.32 | — | — | — | — | — | — | |
| C2Dre-implemented=true2021.11 | 76.64 | — | — | — | 92.32 | — | — | — | — | — | — | |
| CoDiM-bare2021.11 | 76.6 | — | — | — | 92.36 | — | — | — | — | — | — | |
| CoDiMvariant=bare2021.11 | 76.6 | — | — | — | 92.36 | — | — | — | — | — | — | |
| CoDiM-Sup2021.11 | 76.52 | — | — | — | 91.96 | — | — | — | — | — | — | |
| CoDiMvariant=sup2021.11 | 76.52 | — | — | — | 91.96 | — | — | — | — | — | — | |
| HomoscedasticBackbone=ResNet-152, Epochs=902021.05 | 76.5 | — | — | — | 93 | — | — | 0.98 | — | — | — | |
| MoPro2022.03 | 76.31 | — | — | — | — | — | — | — | — | — | — | |
| ResNet-50Throughput=536, Note=baseline2020.01 | 76.3 | — | — | — | — | 76 | 57.7 | — | — | — | — | |
| SSR+2021.11 | 75.76 | — | — | — | 91.76 | — | — | — | — | — | — | |
| CMW-Net-SLBackbone=ResNet-502022.02 | 75.72 | — | — | — | 92.52 | — | — | — | — | — | — | |
| DM-AugDescimplementation=re-implemented2021.11 | 75.52 | — | — | — | 92.12 | — | — | — | — | — | — | |
| DM-AugDescre-implemented=true2021.11 | 75.52 | — | — | — | 92.12 | — | — | — | — | — | — | |
| GJS2021.11 | 75.5 | — | — | — | 91.27 | — | — | — | — | — | — | |
| GJS2021.11 | 75.5 | — | — | — | 91.27 | — | — | — | — | — | — | |
| Robust LRnumber of classes=502021.12 | 75.48 | — | — | — | 93.76 | — | — | — | — | — | — | |
| TCLBackbone=ResNet-182023.03 | 75.4 | — | — | — | 92.4 | — | — | — | — | — | — | |
| UNICONtrained_on=WebVision2022.03 | 75.29 | — | — | — | 93.72 | — | — | — | — | — | — | |
| DivideMix2021.11 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMix2021.11 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMix2022.03 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMixtrained_on=WebVision2022.03 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMixnumber of classes=502021.12 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMixBackbone=ResNet-182023.03 | 75.2 | — | — | — | 90.8 | — | — | — | — | — | — | |
| DivideMixBackbone=Inception-ResNet-v22022.02 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMixtrained_on=WebVision2023.01 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMix2021.11 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| DivideMix2021.08 | 75.2 | — | — | — | 90.84 | — | — | — | — | — | — | |
| Knockoffs-SPRtrained_on=WebVision2023.01 | 74.72 | — | — | — | 92.88 | — | — | — | — | — | — | |
| NGC2022.03 | 74.44 | — | — | — | 91.04 | — | — | — | — | — | — | |
| NGC2021.11 | 74.44 | — | — | — | 91.04 | — | — | — | — | — | — | |
| NGC2021.08 | 74.44 | — | — | — | 91.04 | — | — | — | — | — | — | |
| DivideMixBackbone=ResNet-502022.02 | 74.42 | — | — | — | 91.21 | — | — | — | — | — | — | |
| NGCBackbone=ResNet-182023.03 | 74.4 | — | — | — | 91 | — | — | — | — | — | — | |
| MOITBackbone=ResNet-182023.03 | 73.8 | — | — | — | 91.7 | — | — | — | — | — | — | |
| Teacher: ResNet-34Model=ResNet-342020.11 | 73.31 | — | — | — | — | — | — | — | 3.56 | — | — | |
| ProtoMix2022.03 | 73.3 | — | — | — | 91.2 | — | — | — | — | — | — | |
| RRLBackbone=ResNet-182023.03 | 73.3 | — | — | — | 91.2 | — | — | — | — | — | — | |
| RRL2021.11 | 73.3 | — | — | — | 91.2 | — | — | — | — | — | — | |
| MentorMixData=Mini, Extra clean labels during training=false2019.11 | 72.9 | — | — | — | 91.1 | — | — | — | — | — | — | |
| SPRtrained_on=WebVision2023.01 | 72.32 | — | — | — | 90.92 | — | — | — | — | — | — | |
| PAD-L2Teacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=90 + 30, Training strategy=Distributed (3 GPUs), Scaling rule=Original batch size2020.11 | 71.71 | — | — | — | — | — | — | — | 1.96 | 2,834 | — | |
| FTTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=91, Training strategy=Single GPU2020.11 | 71.56 | — | — | — | — | — | — | — | 1.81 | 5,506 | — | |
| KDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Distributed (3 GPUs), Scaling rule=Linear scaling [7]2020.11 | 71.37 | — | — | — | — | — | — | — | 1.62 | 2,307 | — | |
| KDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Single GPU2020.11 | 71.23 | — | — | — | — | — | — | — | 1.48 | 6,004 | — | |
| ELRInitialization=Uns-CL2022.03 | 71.23 | — | — | — | 88.23 | — | — | — | — | — | — | |
| FTTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=91, Training strategy=Distributed (3 GPUs), Scaling rule=Linear scaling [7]2020.11 | 71.13 | — | — | — | — | — | — | — | 1.38 | 2,215 | — | |
| L2Teacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=90, Training strategy=Distributed (3 GPUs), Scaling rule=Original batch size2020.11 | 71.08 | — | — | — | — | — | — | — | 1.33 | 2,125 | — | |
| CRDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Distributed (3 GPUs), Scaling rule=Original batch size2020.11 | 70.93 | — | — | — | — | — | — | — | 1.18 | 17,912 | — | |
| ATTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Single GPU2020.11 | 70.9 | — | — | — | — | — | — | — | 1.15 | 5,907 | — | |
| CRDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Single GPU2020.11 | 70.81 | — | — | — | — | — | — | — | 1.06 | 35,631 | — | |
| ATTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=100, Training strategy=Distributed (3 GPUs), Scaling rule=Linear scaling [7]2020.11 | 70.55 | — | — | — | — | — | — | — | 0.8 | 2,311 | — | |
| Tf-KDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=90, Training strategy=Single GPU2020.11 | 70.52 | — | — | — | — | — | — | — | 0.77 | 4,634 | — | |
| ELR+2021.11 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELR+2021.11 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELR+2022.03 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELRtrained_on=WebVision2022.03 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELRBackbone=Inception-ResNet-v22022.02 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELR+2021.11 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| ELR+2021.08 | 70.29 | — | — | — | 89.76 | — | — | — | — | — | — | |
| Tf-KDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=90, Training strategy=Distributed (3 GPUs), Scaling rule=Linear scaling [7]2020.11 | 70.21 | — | — | — | — | — | — | — | 0.46 | 1,850 | — | |
| SSKDTeacher backbone=ResNet-34, Student backbone=ResNet-18, # Epochs=130, Training strategy=Distributed (3 GPUs), Scaling rule=Original batch size2020.11 | 70.09 | — | — | — | — | — | — | — | 0.34 | 11,312 | — | |
| Student: ResNet-18Model=ResNet-182020.11 | 69.75 | — | — | — | — | — | — | — | 0 | — | — | |
| Resnet-18Params=11.7M, Evaluation protocol=single crop2017.03 | 68.9 | — | — | — | 88.8 | — | — | — | — | — | — | |
| ELR2022.03 | 68.71 | — | — | — | 87.84 | — | — | — | — | — | — | |
| Scat + Resnet-10Params=12.8M, Evaluation protocol=single crop2017.03 | 68.7 | — | — | — | 88.6 | — | — | — | — | — | — | |
| ELRBackbone=ResNet-182023.03 | 68.7 | — | — | — | 87.8 | — | — | — | — | — | — | |
| VGG-16Params=138M, Evaluation protocol=single crop2017.03 | 68.5 | — | — | — | 88.7 | — | — | — | — | — | — | |
| Saxena et al. (2019)Data=Full, Extra clean labels during training=false2019.11 | 67.5 | — | — | — | — | — | — | — | — | — | — | |
| MentorMixData=Full, Extra clean labels during training=false2019.11 | 67.5 | — | — | — | 87.2 | — | — | — | — | — | — | |
| CMW-NetBackbone=ResNet-502022.02 | 66.44 | — | — | — | 87.68 | — | — | — | — | — | — | |
| MW-NetBackbone=ResNet-502022.02 | 65.8 | — | — | — | 87.52 | — | — | — | — | — | — | |
| Guo et al. (2018)Data=Full, Extra clean labels during training=true2019.11 | 64.8 | — | — | — | 84.9 | — | — | — | — | — | — | |
| MentorNet (2018)Data=Full, Extra clean labels during training=true2019.11 | 64.2 | — | — | — | 84.8 | — | — | — | — | — | — | |
| MentorNet (2018)Data=Mini, Extra clean labels during training=false2019.11 | 63.8 | — | — | — | 85.8 | — | — | — | — | — | — | |
| VanillaData=Full, Extra clean labels during training=false2019.11 | 61.7 | — | — | — | 82.4 | — | — | — | — | — | — | |
| INCVBackbone=Inception-ResNet v22019.05 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Chen et al. (2019)Data=Mini, Extra clean labels during training=false2019.11 | 61.6 | — | — | — | 85 | — | — | — | — | — | — | |
| Iterative-CV2021.11 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Iterative-CV2021.11 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Iterative-CV2022.03 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Iterative-CVtrained_on=WebVision2022.03 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Iterative-CVnumber of classes=502021.12 | 61.6 | — | — | — | 84.98 | — | — | — | — | — | — | |
| Iterative-CVBackbone=ResNet-182023.03 | 61.6 | — | — | — | 84.9 | — | — | — | — | — | — |