Brian Broll
53 linhas
2.0 KiB
Lua
53 linhas
2.0 KiB
Lua
require 'nn'
|
|
-- thanks to https://github.com/soumith/imagenet-multiGPU.torch for this example
|
|
nGPU = 4
|
|
nClasses = 10
|
|
-- from https://code.google.com/p/cuda-convnet2/source/browse/layers/layers-imagenet-1gpu.cfg
|
|
-- this is AlexNet that was presented in the One Weird Trick paper. http://arxiv.org/abs/1404.5997
|
|
local features = nn.Sequential()
|
|
features:add(nn.SpatialConvolution(3,64,11,11,4,4,2,2)) -- 224 -> 55
|
|
features:add(nn.SpatialBatchNormalization(64,1e-3))
|
|
features:add(nn.ReLU(true))
|
|
features:add(nn.SpatialMaxPooling(3,3,2,2)) -- 55 -> 27
|
|
features:add(nn.SpatialConvolution(64,192,5,5,1,1,2,2)) -- 27 -> 27
|
|
features:add(nn.SpatialBatchNormalization(192,1e-3))
|
|
features:add(nn.ReLU(true))
|
|
features:add(nn.SpatialMaxPooling(3,3,2,2)) -- 27 -> 13
|
|
features:add(nn.SpatialConvolution(192,384,3,3,1,1,1,1)) -- 13 -> 13
|
|
features:add(nn.SpatialBatchNormalization(384,1e-3))
|
|
features:add(nn.ReLU(true))
|
|
features:add(nn.SpatialConvolution(384,256,3,3,1,1,1,1)) -- 13 -> 13
|
|
features:add(nn.SpatialBatchNormalization(256,1e-3))
|
|
features:add(nn.ReLU(true))
|
|
features:add(nn.SpatialConvolution(256,256,3,3,1,1,1,1)) -- 13 -> 13
|
|
features:add(nn.SpatialBatchNormalization(256,1e-3))
|
|
features:add(nn.ReLU(true))
|
|
features:add(nn.SpatialMaxPooling(3,3,2,2)) -- 13 -> 6
|
|
|
|
-- features:cuda()
|
|
-- features = makeDataParallel(features, nGPU) -- defined in util.lua
|
|
|
|
local classifier = nn.Sequential()
|
|
classifier:add(nn.View(256*6*6))
|
|
|
|
classifier:add(nn.Dropout(0.5))
|
|
classifier:add(nn.Linear(256*6*6, 4096))
|
|
classifier:add(nn.BatchNormalization(4096, 1e-3))
|
|
classifier:add(nn.ReLU())
|
|
|
|
classifier:add(nn.Dropout(0.5))
|
|
classifier:add(nn.Linear(4096, 4096))
|
|
classifier:add(nn.BatchNormalization(4096, 1e-3))
|
|
classifier:add(nn.ReLU())
|
|
|
|
classifier:add(nn.Linear(4096, nClasses))
|
|
classifier:add(nn.LogSoftMax())
|
|
|
|
-- classifier:cuda()
|
|
|
|
local model = nn.Sequential():add(features):add(classifier)
|
|
model.imageSize = 256
|
|
model.imageCrop = 224
|
|
|
|
return model
|