-- thanks to https://github.com/soumith/imagenet-multiGPU.torch for this example
require 'nn'
nClasses = 7
local features = nn.Sequential()

features:add(nn.SpatialConvolution(3, 96, 11, 11, 4, 4))
features:add(nn.ReLU(true))
features:add(nn.SpatialMaxPooling(2, 2, 2, 2))

features:add(nn.SpatialConvolution(96, 256, 5, 5, 1, 1))
features:add(nn.ReLU(true))
features:add(nn.SpatialMaxPooling(2, 2, 2, 2))

features:add(nn.SpatialConvolution(256, 512, 3, 3, 1, 1, 1, 1))
features:add(nn.ReLU(true))

features:add(nn.SpatialConvolution(512, 1024, 3, 3, 1, 1, 1, 1))
features:add(nn.ReLU(true))

features:add(nn.SpatialConvolution(1024, 1024, 3, 3, 1, 1, 1, 1))
features:add(nn.ReLU(true))
features:add(nn.SpatialMaxPooling(2, 2, 2, 2))

-- features:cuda()
-- features = makeDataParallel(features, nGPU) -- defined in util.lua

-- 1.3. Create Classifier (fully connected layers)
local classifier = nn.Sequential()
classifier:add(nn.View(1024*5*5))
classifier:add(nn.Dropout(0.5))
classifier:add(nn.Linear(1024*5*5, 3072))
classifier:add(nn.Threshold(0, 0.000001))

classifier:add(nn.Dropout(0.5))
classifier:add(nn.Linear(3072, 4096))
classifier:add(nn.Threshold(0, 0.000001))

classifier:add(nn.Linear(4096, nClasses))
classifier:add(nn.LogSoftMax())

-- classifier:cuda()

-- 1.4. Combine 1.2 and 1.3 to produce final model
local model = nn.Sequential():add(features):add(classifier)
model.imageSize = 256
model.imageCrop = 224
return model