Norm matters

This repository holds code to reproduce results in the paper: "Norm matters: efficient and accurate normalization schemes in deep networks"

---

This is a complete training example for Deep Convolutional Networks on various datasets (ImageNet, Cifar10, Cifar100, MNIST).

It is based off imagenet example in pytorch with some helpful additions such as:

• Training on several datasets other than imagenet
• Complete logging of trained experiment
• Graph visualization of the training/validation loss and accuracy
• Definition of preprocessing and optimization regime for each model

Dependencies

• pytorch
• torchvision to load the datasets, perform image transforms
• pandas for logging to csv
• bokeh for training visualization

Data

• Configure your dataset path at data.py.
• To get the ILSVRC data, you should register on their site for access: http://www.image-net.org/

Model configuration

Network model is defined by writing a .py file in models folder, and selecting it using the model flag. Model function must be registered in models/__init__.py The model function must return a trainable network. It can also specify additional training options such optimization regime (either a dictionary or a function), and input transform modifications.

e.g for a model definition:

class Model(nn.Module):

    def __init__(self, num_classes=1000):
        super(Model, self).__init__()
        self.model = nn.Sequential(...)

        self.regime = [
            {'epoch': 0, 'optimizer': 'SGD', 'lr': 1e-2,
                'weight_decay': 5e-4, 'momentum': 0.9},
            {'epoch': 15, 'lr': 1e-3, 'weight_decay': 0}
        ]

        self.input_transform = {
            'train': transforms.Compose([...]),
            'eval': transforms.Compose([...])
        }
    def forward(self, inputs):
        return self.model(inputs)
        
 def model(**kwargs):
        return Model()