Please star ModelFeast if it helps you. This is very important to me! Thanks very much !
ModelFeast is more than model-zoo! It is:
- A gather of the most popular 2D, 3D CNN models
- A tool to make deep learn much more simply and flexibly
- A pytorch project template
- Xception
- InceptionV3
- InceptionResnetV2
- SqueezeNet1_0, SqueezeNet1_1
- VGG11, VGG13, VGG16, VGG19
- ResNet18, ResNet34, ResNet50, ResNet101, ResNet152
- ResNext101_32x4d, ResNext101_64x4d
- DenseNet121, DenseNet169, DenseNet201, DenseNet161
Pretrained models ( trained on ImageNet ) for 2D CNN is now avalible on Baiduyun(fst6) and Google Drive
- 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
This part is still on progress. Not avalible to train now, but model architecture can been seen here.
Determine what you need and read corresponding tutorials
- I want to train a model as simple as possible
- I just need the codes of CNNs
- I need a standard pytorch project template
Or you can use modelfeast simply via pip !
pip3 install modelfeast
The features are more than you could think of:
- Train and save model within 3 lines !
- Multi GPU support !
- Include the most popular 2D CNN, 3D CNN, and CRNN models !
- Allow any input image size (pytorch official model zoo limit your input size harshly) !
- Help you sweep all kinds of classification competitions.
https://github.com/lanpa/tensorboardX
https://github.com/pytorch/vision/tree/master/torchvision/models
https://github.com/kenshohara/3D-ResNets-PyTorch
https://github.com/victoresque/pytorch-template