Reproduce ResNet-v3(Aggregated Residual Transformations for Deep Neural Network)
-
1-crop validation error on ImageNet (center 224x224 crop from resized image with shorter side=256):
model top-1 top-5 VGG-16 28.5% 9.9% ResNet-50 24.7% 7.8% ResNeXt-50 23.1% 6.7% ResNet-101 23.6% 7.1% ResNeXt-101(32x4d) 22.1% 5.8% ResNet-152 23.0% 6.7% ResNeXt-152 21.3% 5.6% ResNeXt-101(64x4d) 20.8% 5.5% -
1-crop validation error on ImageNet (center 320x320 crop from resized image with shorter side=320):
Network | crop-size | top-1 | top-5 |
---|---|---|---|
ResNet-152-v1 | 320x320 | 21.3% | 5.5% |
ResNet-152-v2 | 320x320 | 21.1% | 5.5% |
ResNet-200-v1 | 320x320 | 21.8% | 6.0% |
ResNet-200-v2 | 320x320 | 20.7% | 5.3% |
ResNeXt-50 | 320x320 | 21.9% | 5.9% |
ResNeXt-101(32x4d) | 320x320 | 20.2% | 4.9% |
ResNeXt-152 | 320x320 | 19.9% | 4.8% |
ResNeXt-101(64x4d) | 320x320 | 19.5% | 4.8% |
ResNeXt-50 OneDrive download: link
ResNeXt-101 OneDrive download: link
ResNeXt-152 OneDrive download: link