Convolutional networks using Torch

This is a complete training example for Deep Convolutional Networks on various datasets (ImageNet, Cifar10, Cifar100, STL10, SVHN, MNIST).

It uses TorchNet (https://github.com/torchnet/torchnet) for fast data loading and measurements.

It aims to replace both https://github.com/eladhoffer/ConvNet-torch and https://github.com/eladhoffer/ImageNet-Training

Multiple GPUs are also supported by using nn.DataParallelTable (https://github.com/torch/cunn/blob/master/docs/cunnmodules.md).

Dependencies

• Torch (http://torch.ch)
• torchnet (https://github.com/torchnet/torchnet)
• cudnn.torch (https://github.com/soumith/cudnn.torch) for faster training (optional)

Data

• To get Cifar data, use @soumith's repo: https://github.com/soumith/cifar.torch.git
• To get the ILSVRC data, you should register on their site for access: http://www.image-net.org/
• All data related functions used for training are available at data.lua.

Model configuration

Network model is defined by writing a

.lua file in models folder, and selecting it using the model flag. The model file must return a trainable network. It can also specify additional training options such optimization regime, input size modifications.

e.g for a model file:

local model = nn.Sequential():add(...)
  model.inputSize = 224
  model.reshapeSize = 256
  model.regime = {
    epoch        = {1,    19,   30,   44,   53  },
    learningRate = {1e-2, 5e-3, 1e-3, 5e-4, 1e-4},
    weightDecay  = {5e-4, 5e-4, 0,    0,    0   }
  }
return model

Training

You can start training using main.lua by typing:

th main.lua -model AlexNet -LR 0.01

or if you have 2 gpus availiable,

th main.lua -model AlexNet -LR 0.01 -nGPU 2 -batchSize 256

A more elaborate example continuing a pretrained network and saving intermediate results

th main.lua -model GoogLeNet_BN -batchSize 64 -nGPU 2 -save GoogLeNet_BN -bufferSize 9600 -LR 0.01 -checkpoint 320000 -weightDecay 1e-4 -load ./pretrainedNet.t7

Output

Training output will be saved to folder defined with save flag.

Additional flags