szagoruyko

local SpatialAffine, parent = torch.class('nn.SpatialAffine', 'nn.Module')

function SpatialAffine:__init(nOutput)
  parent.__init(self)
  self.weight = torch.Tensor(nOutput)
  self.bias = torch.Tensor(nOutput)
end

function SpatialAffine:updateOutput(input)
  local nFeature = self.weight:numel()

szagoruyko

require 'loadcaffe'
require 'optim'
local matio = require 'matio'

local dataset = torch.load('/opt/datasets/tiny-datasets/cifar10_whitened.t7')

local net = loadcaffe.load('/tmp/e56253735ef32c3c296d/train_val.prototxt','./cifar10_nin.caffemodel')
net:evaluate()
print(net)

szagoruyko

szagoruyko

Network-in-Network trained in Torch7

Trained by me using https://github.com/soumith/imagenet-multiGPU.torch, achieves 62.6% top1 center crop accuracy on ImageNet validation set. Tested here: https://github.com/szagoruyko/imagenet-validation.torch

Download links:

1. https://www.dropbox.com/s/mclw90yba6eml60/nin_bn_final.t7 (31 MB)
2. https://www.dropbox.com/s/npmr5egvjbg7ovb/nin_nobn_final.t7 (31 MB) - Batch Normalization intergrated into convolutional layers

szagoruyko

AlexNet trained in Torch7

Trained by me using https://github.com/soumith/imagenet-multiGPU.torch, achieves 56.7% top1 center crop accuracy on ImageNet validation set. Tested here: https://github.com/szagoruyko/imagenet-validation.torch

Download link: https://yadi.sk/d/uX6id_yZoU8FC (247 MB)

Load as:

szagoruyko

ts = {
  model = net,
  unpack = function(self)
    for k,v in ipairs(self.model:listModules()) do
      if v.weight and not v.gradWeight then
        v.gradWeight = v.weight:clone()
        v.gradBias = v.bias:clone()
      end
    end
    return self.model

szagoruyko

function nn.Module:replace(callback)
  local out = callback(self)
  if self.modules then
    for i, module in ipairs(self.modules) do
      self.modules[i] = module:replace(callback)
    end
  end
  return out
end

szagoruyko

-- Usage:
-- find . -name model.net -print0 | xargs -0 -n 1 th ~/clearState.lua

require 'cudnn'

local name = arg[1]

assert(paths.filep(name))

print'before'

szagoruyko

require 'cv.highgui'
require 'cv.videoio'
require 'cv.imgproc'
require 'nn'
--require 'clnn'

local cap = cv.VideoCapture{device=0}
if not cap:isOpened() then
  print("Failed to open the default camera")
  os.exit(-1)

szagoruyko

require 'nn'

-- you just need to provide the linear module you want to convert,
-- and the dimensions of the field of view of the linear layer
function convertLinear2Conv1x1(linmodule,in_size)
   local s_in = linmodule.weight:size(2)/(in_size[1]*in_size[2])
   local s_out = linmodule.weight:size(1)
   local convmodule = nn.SpatialConvolutionMM(s_in,s_out,in_size[1],in_size[2],1,1)
   convmodule.weight:copy(linmodule.weight)
   convmodule.bias:copy(linmodule.bias)