Deep Learning on ImageNet using Torch

This is a complete training example for Deep Convolutional Networks on the ILSVRC classification task.

Data is preprocessed and cached as a LMDB data-base for fast reading. A separate thread buffers images from the LMDB record in the background.

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

This code allows training at 4ms/sample with the AlexNet model and 2ms for testing on a single GPU (using Titan Z with 1 active gpu)

##Dependencies

• "eladtools" (https://github.com/eladhoffer/eladtools) for DataProvider class and optimizer.
• “lmdb.torch” (http://github.com/eladhoffer/lmdb.torch) for LMDB usage.
• “cudnn.torch” (https://github.com/soumith/cudnn.torch) for faster training. Can be avoided by changing "cudnn" to "nn" in models.

##Data

• To get the ILSVRC data, you should register on their site for access: http://www.image-net.org/
• Configure the data location and save dir in Config.lua.
• LMDB records for fast read access are created by running CreateLMDBs.lua. Saving the default 256x256 image for each sample will result with a ~350GB file for training data, and a ~13.2GB file for validation data.

##Training You can start training using:

th Main.lua -network AlexNet -LR 0.01

or if you have 2 gpus availiable,

th Main.lua -network AlexNet -LR 0.01 -nGPU 2 -batchSize 256

Buffer size should be adjusted to suit the used hardware and configuration. Default value is 1280 (10 batches of 128) which works well when using a non SSD drive and 1 GPU.

##Additional flags

Flag	Default Value	Description
modelsFolder	./Models/	Models Folder
network	AlexNet	Model file - must return valid network.
LR	0.01	learning rate
LRDecay	0	learning rate decay (in # samples
weightDecay	5e-4	L2 penalty on the weights
momentum	0.9	momentum
batchSize	128	batch size
optimization	sgd	optimization method
epoch	-1	number of epochs to train (-1 for unbounded)
threads	8	number of threads
type	cuda	float or cuda
bufferSize	1280	buffer size
devid	1	device ID (if using CUDA)
nGPU	1	num of gpu devices used
load	none	load existing net weights
save	time-identifier	save directory
shuffle	false	shuffle training samples
augment	true	Augment training data