I have builded the most popular CNN in pytorch, they are slightly modifed to be able to accept any input image size.
You can find a list of models ./models/__init__.py
- Xception
- InceptionV3
- InceptionResnetV2
- SqueezeNet1_0, SqueezeNet1_1
- VGG11, VGG13, VGG16, VGG19
- ResNet18, ResNet34, ResNet50, ResNet101, ResNet152
- ResNext101_32x4d, ResNext101_64x4d
- DenseNet121, DenseNet169, DenseNet201, DenseNet161
3D CNN are mainly used in video, medical image etc.
- resnet18v2_3d, resnet34v2_3d, resnet50v2_3d, resnet101v2_3d, resnet152v2_3d, resnet200v2_3d
- resnext50_3d, resnext101_3d, resnext152_3d
- densenet121_3d, densenet169_3d, densenet201_3d, densenet264_3d
- resnet10_3d, resnet18_3d, resnet34_3d, resnet101_3d, resnet152_3d, resnet200_3d
- wideresnet50_3d
- i3d50, i3d101, i3d152
I easily achieve 9th in the medical image classification competition with DenseNet201_3d in ModelFeast!
CNN-RNN are mainly used in video, medical image etc. This part is still on progress. Not avalible to train now, but model architecture can been seen here.