Image Classification on ImageNet (val) (Accuracy and Timing)
15.42Top-1 ErrorColornet-121
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Colornet-121testing_mode=10-crop2019.02 | 15.42 | 3.89 | — | — | |
| Colornet-121testing_mode=single-crop2019.02 | 17.65 | 5.22 | — | — | |
| SKNet-101Crop Size=320x320, #P=48.9M, GFLOPS=8.462019.03 | 18.4 | — | — | — | |
| SE-ShuffleNet v2-164FLOPs=12.7G, with residual=true2018.07 | 18.56 | — | — | — | |
| SENet-101Crop Size=320x320, #P=49.2M, GFLOPS=82019.03 | 18.61 | — | — | — | |
| SENetFLOPs=20.7G2018.07 | 18.68 | — | — | — | |
| DPN-98Crop Size=320x320, #P=61.6M, GFLOPS=11.72019.03 | 18.9 | — | — | — | |
| ResNeXt-101 + BAMCrop Size=320x320, #P=44.6M, GFLOPS=8.052019.03 | 19.15 | — | — | — | |
| DPN-92Crop Size=320x320, #P=37.7M, GFLOPS=6.52019.03 | 19.3 | — | — | — | |
| SKNet-50Crop Size=320x320, #P=27.5M, GFLOPS=4.472019.03 | 19.32 | — | — | — | |
| ResNeXt-101 + CBAMCrop Size=320x320, #P=49.2M, GFLOPS=82019.03 | 19.42 | — | — | — | |
| Attention-92Crop Size=320x320, #P=51.3M, GFLOPS=10.432019.03 | 19.5 | — | — | — | |
| SENet-50Crop Size=320x320, #P=27.7M, GFLOPS=4.252019.03 | 19.71 | — | — | — | |
| ResNeXt-101Crop Size=320x320, #P=44.3M, GFLOPS=7.992019.03 | 19.86 | — | — | — | |
| Inception-ResNetV2Crop Size=320x320, #P=55.0M, GFLOPS=13.222019.03 | 19.9 | — | — | — | |
| InceptionV4Crop Size=320x320, #P=42.0M, GFLOPS=12.312019.03 | 20 | — | — | — | |
| ResNeXt-50 + BAMCrop Size=320x320, #P=25.4M, GFLOPS=4.312019.03 | 20.15 | — | — | — | |
| SKNet-101Number of parameters=48.9M, Crop size=224x2242019.03 | 20.19 | — | — | — | |
| SKNet-101Crop Size=224x224, #P=48.9M, GFLOPS=8.462019.03 | 20.19 | — | — | — | |
| DPN-98Number of parameters=61.57M, Cardinality x Bottleneck Width=32×4d, Crop size=224x2242019.03 | 20.2 | — | — | — | |
| DPN-98Crop Size=224x224, #P=61.6M, GFLOPS=11.72019.03 | 20.2 | — | — | — | |
| ResNeXt-50 + CBAMCrop Size=320x320, #P=27.7M, GFLOPS=4.252019.03 | 20.38 | — | — | — | |
| SENet-101Number of parameters=49.2M, Crop size=224x2242019.03 | 20.58 | — | — | — | |
| SENet-101Crop Size=224x224, #P=49.2M, GFLOPS=82019.03 | 20.58 | — | — | — | |
| ResNeXt-101 + CBAMCrop Size=224x224, #P=49.2M, GFLOPS=82019.03 | 20.6 | — | — | — | |
| ResNeXt-101 + BAMCrop Size=224x224, #P=44.6M, GFLOPS=8.052019.03 | 20.67 | — | — | — | |
| DPN-92Number of parameters=37.67M, Cardinality x Bottleneck Width=32×3d, Crop size=224x2242019.03 | 20.7 | — | — | — | |
| DPN-92Crop Size=224x224, #P=37.7M, GFLOPS=6.52019.03 | 20.7 | — | — | — | |
| SKNet-50Number of parameters=27.5M, Crop size=224x2242019.03 | 20.79 | — | — | — | |
| SKNet-50Crop Size=224x224, #P=27.5M, GFLOPS=4.472019.03 | 20.79 | — | — | — | |
| DenseNet-264testing_mode=10-crop2019.02 | 20.8 | 5.29 | — | — | |
| ResNeXt-50Crop Size=320x320, #P=25.0M, GFLOPS=4.242019.03 | 21.05 | — | — | — | |
| ResNeXt-101Crop Size=224x224, #P=44.3M, GFLOPS=7.992019.03 | 21.11 | — | — | — | |
| SENet-50Number of parameters=27.7M, Crop size=224x2242019.03 | 21.12 | — | — | — | |
| SENet-50Crop Size=224x224, #P=27.7M, GFLOPS=4.252019.03 | 21.12 | — | — | — | |
| ResNeXt-101Number of parameters=44.30M, Cardinality x Bottleneck Width=32×4d, Crop size=224x2242019.03 | 21.2 | — | — | — | |
| Inception V3Crop Size=320x320, #P=27.1M, GFLOPS=5.732019.03 | 21.2 | — | — | — | |
| DenseNet-232Number of parameters=55.80M, Growth rate (k)=48, Crop size=224x2242019.03 | 21.3 | — | — | — | |
| ResNeXt-50 + CBAMCrop Size=224x224, #P=27.7M, GFLOPS=4.252019.03 | 21.4 | — | — | — | |
| DenseNet-201testing_mode=10-crop2019.02 | 21.46 | 5.54 | — | — | |
| ResNet-152Number of parameters=60.19M, Crop size=224x2242019.03 | 21.7 | — | — | — | |
| ResNeXt-50 + BAMCrop Size=224x224, #P=25.4M, GFLOPS=4.312019.03 | 21.7 | — | — | — | |
| AttentionNeXt-56Crop Size=224x224, #P=31.9M, GFLOPS=6.322019.03 | 21.76 | — | — | — | |
| DenseNet-264testing_mode=single-crop2019.02 | 22.15 | 6.12 | — | — | |
| DenseNet-264Number of parameters=33.34M, Growth rate (k)=32, Crop size=224x2242019.03 | 22.2 | — | — | — | |
| ResNeXt-50Number of parameters=25.00M, Cardinality x Bottleneck Width=32×4d, Crop size=224x2242019.03 | 22.2 | — | — | — | |
| ResNeXt-50Crop Size=224x224, #P=25.0M, GFLOPS=4.242019.03 | 22.23 | — | — | — | |
| DenseNet-201testing_mode=single-crop2019.02 | 22.58 | 6.34 | — | — | |
| ResNet-101Number of parameters=44.55M, Crop size=224x2242019.03 | 22.6 | — | — | — | |
| DenseNet-201Number of parameters=20.01M, Growth rate (k)=32, Crop size=224x2242019.03 | 22.6 | — | — | — | |
| SKNet-26Number of parameters=16.8M, Crop size=224x2242019.03 | 22.74 | — | — | — | |
| ShuffleNet v2-50FLOPs=2.3G2018.07 | 22.8 | — | — | — | |
| Batch Normalization (BN)Backbone=ResNet-50, Batch size=32 images/GPU2018.03 | 23.6 | — | — | — | |
| DenseNet-121testing_mode=10-crop2019.02 | 23.61 | 6.66 | — | — | |
| DPN-68Number of parameters=12.61M, Cardinality x Bottleneck Width=32×4d, Crop size=224x2242019.03 | 23.7 | — | — | — | |
| DenseNet-169Number of parameters=14.15M, Growth rate (k)=32, Crop size=224x2242019.03 | 23.8 | — | — | — | |
| ResNet-50Number of parameters=25.56M, Crop size=224x2242019.03 | 23.9 | — | — | — | |
| ResNet-50FLOPs=3.8G2018.07 | 24 | — | — | — | |
| Group Normalization (GN)Backbone=ResNet-50, Batch size=32 images/GPU2018.03 | 24.1 | — | — | — | |
| DenseNet-121testing_mode=single-crop2019.02 | 25.02 | 7.71 | — | — | |
| ShuffleNet v1-50FLOPs=2.3G, implementation=our impl.2018.07 | 25.2 | — | — | — | |
| Layer Normalization (LN)Backbone=ResNet-50, Batch size=32 images/GPU2018.03 | 25.3 | — | — | — | |
| Weight Normalization (WN)Backbone=ResNet-50, Batch size=32 images/GPU2018.03 | 28.2 | — | — | — | |
| Instance Normalization (IN)Backbone=ResNet-50, Batch size=32 images/GPU2018.03 | 28.4 | — | — | — | |
| iSQRT-COVBackbone=AlexNet, normalization=trace2017.12 | 38.45 | 17.52 | 2.55 | 0.81 | |
| MPN-COVBackbone=AlexNet2017.12 | 38.51 | 17.6 | 3.89 | 2.59 | |
| G2DeNetBackbone=AlexNet2017.12 | 38.71 | 17.66 | 9.86 | 5.88 | |
| iSQRT-COVBackbone=AlexNet, normalization=Frobenius norm2017.12 | 38.78 | 17.67 | 2.56 | 0.81 | |
| B-CNNBackbone=AlexNet2017.12 | 39.89 | 18.32 | 1.92 | 0.83 | |
| Improved B-CNNBackbone=AlexNet, post-processing=matrix square root, element-wise square root and l2 normalizations2017.12 | 40.75 | 18.91 | 15.48 | 13.04 | |
| AlexNetBackbone=AlexNet2017.12 | 41.8 | 19.2 | 1.32 | 0.77 | |
| DeepO2PBackbone=AlexNet2017.12 | 42.16 | 19.62 | 11.23 | 7.04 |