- More learning rate decay strategies (currently only dropping at 1/2 and 3/4 of the epochs)
- CPU support
- SVHN-small (without extra training data)
- SVHN
- MNIST
- ImageNet
- Comparing tensorboard_logger v.s. pycrayon
- Adding acknowledgement
- Custom models & criterions tutorial
- Custom train & test functions tutorial
- Custom datasets tutorial
- Custom initialization
- Adding an example project killing scenario
- Adding license
- Pretrained models
- Iteration mode (Counting iterations instead of epochs)
- Pep8 check
- AlexNet (paper)
- VGGNet (paper)
- SqueezeNet (paper) (code)
- Pre-ResNet (paper) (code)
- Pre-ResNet with Stochastic Depth
- Wide ResNet (paper) (code)
- DenseNet (paper) (code) (Our implementation is buggy now, we encourage you to put Andreas Veit's implementation or Brandon Amos's implementation into models folder and add a
createModelfunction to it) - PyramidalNet (paper)(code)
- PyramidalNet with Separated Stochastic Depth (paper)(code)
- ResNeXt (paper) (code)
- MSDNet (paper) (code)
- Steerable CNN (paper)