Image Classification on ImageNet (Error Rates)
11.39Top-1 ErrorEfficientNet-L2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| EfficientNet-L2Backbone=EfficientNet-L2, SAM Optimization=true, Pre-training Dataset=ImageNet plus JFT2020.10 | 11.39 | — | — | |
| ViT2020.10 | 11.45 | — | — | |
| EfficientNet-L2Backbone=EfficientNet-L2, SAM Optimization=false, Pre-training Dataset=ImageNet plus JFT2020.10 | 11.8 | — | — | |
| KD-forAAPre-training Dataset=ImageNet-only2020.10 | 14.2 | — | — | |
| EfficientNet-b7Backbone=EfficientNet-b7, SAM Optimization=true, Pre-training Dataset=ImageNet-only2020.10 | 15.14 | — | — | |
| EfficientNet-b7Backbone=EfficientNet-b7, SAM Optimization=false, Pre-training Dataset=ImageNet-only2020.10 | 15.3 | — | — | |
| AlignMixup/AEBackbone=ResNet-50, Training Epochs=300, Parameters=35M, msec/batch=6882021.03 | 18.83 | — | — | |
| AlignMixupBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=4502021.03 | 20.68 | — | — | |
| PuzzleMixBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=846, Data Source=Reported by PuzzleMix [32]2021.03 | 21.24 | — | — | |
| SaliencyMixBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=463, Data Source=Reported by Authors [57]2021.03 | 21.26 | — | — | |
| CutMixBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=427, Data Source=Reported by PuzzleMix [32]2021.03 | 21.4 | — | — | |
| ManifoldBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=441, Data Source=Reported by PuzzleMix [32]2021.03 | 22.5 | — | — | |
| InputBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=436, Data Source=Reported by PuzzleMix [32]2021.03 | 22.58 | — | — | |
| BaselineBackbone=ResNet-50, Training Epochs=300, Parameters=25M, msec/batch=4182021.03 | 23.68 | — | — | |
| Classifier OnlyBackbone=ResNet-50, # Models=1, Model Size=94 MB2018.01 | 23.84 | 7.13 | — | |
| PiggybackBackbone=ResNet-50, # Models=1, Model Size=109 MB2018.01 | 23.84 | 7.13 | — | |
| Individual NetworksBackbone=ResNet-50, # Models=6, Model Size=564 MB2018.01 | 23.84 | 7.13 | — | |
| PackNetBackbone=ResNet-50, # Models=1, Model Size=103 MB2018.01 | 24.29 | 7.18 | — | |
| Classifier OnlyBackbone=DenseNet-121, # Models=1, Model Size=28 MB2018.01 | 25.56 | 8.02 | — | |
| PiggybackBackbone=DenseNet-121, # Models=1, Model Size=33 MB2018.01 | 25.56 | 8.02 | — | |
| Individual NetworksBackbone=DenseNet-121, # Models=6, Model Size=168 MB2018.01 | 25.56 | 8.02 | — | |
| PackNetBackbone=DenseNet-121, # Models=1, Model Size=31 MB2018.01 | 25.6 | 7.89 | — | |
| Classifier OnlyBackbone=VGG-16 BN, # Models=1, Model Size=537 MB2018.01 | 26.63 | 8.49 | — | |
| PiggybackBackbone=VGG-16 BN, # Models=1, Model Size=621 MB2018.01 | 26.63 | 8.49 | — | |
| Individual NetworksBackbone=VGG-16 BN, # Models=6, Model Size=3,222 MB2018.01 | 26.63 | 8.49 | — | |
| PackNetBackbone=VGG-16 BN, # Models=1, Model Size=587 MB2018.01 | 27.18 | 8.69 | — | |
| PiggybackBackbone=VGG-16, # Models=1, Size=554 MB2018.01 | 28.42 | 9.61 | — | |
| Individual NetworksBackbone=VGG-16, # Models=2, Size=1,074 MB2018.01 | 28.42 | 9.61 | — | |
| Individual NetworksBackbone=VGG-16, Number of Models=2, Model Size=1,096 MB2017.11 | 28.42 | 9.61 | — | |
| Classifier OnlyBackbone=VGG-16, # Models=1, Size=562 MB2017.11 | 28.42 | 9.61 | — | |
| Individual NetworksBackbone=VGG-16, # Models=4, Size=2,173 MB2017.11 | 28.42 | 9.61 | — | |
| Classifier OnlyBackbone=VGG-16, # Models=1, Size=537 MB2018.01 | 28.42 | 9.61 | — | |
| PiggybackBackbone=VGG-16, # Models=1, Size=621 MB2018.01 | 28.42 | 9.61 | — | |
| Individual NetworksBackbone=VGG-16, # Models=6, Size=3,222 MB2018.01 | 28.42 | 9.61 | — | |
| PackNetBackbone=VGG-16, # Models=1, Size=554 MB2018.01 | 29.33 | 9.99 | — | |
| PackNetBackbone=VGG-16, Pruning Ratio=0.50, Number of Models=1, Model Size=576 MB2017.11 | 29.33 | 9.99 | — | |
| PackNetBackbone=VGG-16, Pruning Ratios=0.50, 0.75, 0.75, # Models=1, Size=595 MB2017.11 | 29.33 | 9.99 | — | |
| PackNetBackbone=VGG-16, # Models=1, Size=587 MB2018.01 | 29.33 | 9.99 | — | |
| 100% SupervisedBackbone=ResNet-18, Supervision Level=100%2019.05 | 30.43 | 10.76 | — | |
| PackNetBackbone=VGG-16, Pruning Ratio=0.75, Number of Models=1, Model Size=576 MB2017.11 | 30.87 | 10.93 | — | |
| PackNetBackbone=VGG-16, Pruning Ratios=0.75, 0.75, 0.75, # Models=1, Size=595 MB2017.11 | 30.87 | 10.93 | — | |
| Jointly Trained NetworkBackbone=VGG-16, # Models=1, Size=537 MB2018.01 | 33.49 | 12.25 | — | |
| Jointly Trained NetworkBackbone=VGG-16, Number of Models=1, Model Size=559 MB2017.11 | 33.49 | 12.25 | — | |
| LwFBackbone=VGG-16, # Models=1, Size=562 MB2017.11 | 39.23 | 16.94 | — | |
| ADA-NetBackbone=ResNet-18, Supervision Level=10%2019.05 | 44.91 | 21.18 | — | |
| Dual-View Deep Co-TrainingBackbone=ResNet-18, Supervision Level=10%2019.05 | 46.5 | 22.73 | — | |
| Mean TeacherBackbone=ResNet-18, Supervision Level=10%2019.05 | 49.07 | 23.59 | — | |
| 10% SupervisedBackbone=ResNet-18, Supervision Level=10%2019.05 | 52.23 | 27.54 | — | |
| MobileNet-224Params=4.2M, Mult-Adds=569M2017.12 | 70.6 | 89.5 | — | |
| ShuffleNet (2x)Params=5M, Mult-Adds=524M2017.12 | 70.9 | 89.8 | — | |
| NASNet-AN=4, F=44, Params=5.3M, Mult-Adds=564M2017.12 | 74 | 91.6 | — | |
| AmoebaNet-BN=3, F=62, Params=5.3M, Mult-Adds=555M2017.12 | 74 | 91.5 | — | |
| PNASNet-5N=3, F=54, Params=5.1M, Mult-Adds=588M2017.12 | 74.2 | 91.9 | — | |
| AmoebaNet-AN=4, F=50, Params=5.1M, Mult-Adds=555M2017.12 | 74.5 | 92 | — | |
| AmoebaNet-CN=4, F=50, Params=6.4M, Mult-Adds=570M2017.12 | 75.7 | 92.4 | — | |
| Colorization (Col)Protocol=Unsupervised classification2018.07 | — | — | 62.5 | |
| Combination of MS + Ex + RP + ColProtocol=Unsupervised classification2018.07 | — | — | 69.3 | |
| CPCProtocol=Unsupervised classification2018.07 | — | — | 73.6 | |
| Exemplar (Ex)Protocol=Unsupervised classification2018.07 | — | — | 53.1 | |
| Motion Segmentation (MS)Protocol=Unsupervised classification2018.07 | — | — | 48.3 | |
| Relative Position (RP)Protocol=Unsupervised classification2018.07 | — | — | 59.2 |