farrajota

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)
--[[
    Convert Linear modules to convolution modules. 
    
    Arguments

farrajota

m.updateGradInput = function(self,i,o) end -- for the gradInput
m.accGradParameters = function(self,i,o) end -- for freezing the parameters

farrajota

local function mysplit(inputstr, sep)
-- split strings by any type of separator
  if sep == nil then
    sep = "%s"
  end
  local t={} ; i=1
  for str in string.gmatch(inputstr, "([^"..sep.."]+)") do
    t[i] = str
    i = i + 1
  end

farrajota

-- Available torch7 pre-trained models for download 

Alexnet (trained by me) 
cudnn: https://ln.sync.com/dl/3d9c28e80#cvjp4mjs-mc2y8jrt-9wq5yebx-yrdchcei
mean = {0.48037518790839, 0.45039056120456, 0.39922636057037}
std =  {0.27660147027775, 0.26883440068399, 0.28014687231841}
img size: 3x224x224

overfeat: https://github.com/jhjin/overfeat-torch
mean = {118.380948, 118.380948, 118.380948}

farrajota

function save_configs(filename, opt)
-- prints the contents of a table into a file

    local function table_print (tt, indent, done)
                    done = done or {}
                    indent = indent or 0
                    if type(tt) == "table" then
                      local sb = {}
                      for key, value in pairs (tt) do
                        table.insert(sb, string.rep (" ", indent)) -- indent it

farrajota

-- multiple learning rates per network. Optimizes two copies of a model network and checks if the optimization steps (2) and (3) produce the same weights/parameters.

require 'torch'
require 'nn'
require 'optim'
torch.setdefaulttensortype('torch.FloatTensor')

-- (1) Define a model for this example.
local model = nn.Sequential()
model:add(nn.Linear(10,20))

farrajota

--[[
    Create a binary tree using two ways: NN containers and nn.gModule containers. 
    This example is fairly simple, and the default fully-connected layers are all
    of size 100. However, this should also be simple to modify to allow different 
    fc layers with varying inputs/outputs if desired (for example: input a table
    storing input+output configuration values for each of the sub-branch's level).
]]

require 'nn'
require 'nngraph'

farrajota

require 'image'

local M = {}

function M.Compose(transforms)
   return function(input)
      for _, transform in ipairs(transforms) do
         input = transform(input)
      end
      return input

farrajota

function mpii_convert_json(  )
% convert mpii annotations .mat file to .json


%% load annotation file
fprintf('Load annotations... ')
data = load('/media/HDD2/Datasets/Human_Pose/mpii/mpii_human_pose_v1_u12_2/mpii_human_pose_v1_u12_1.mat');
fprintf('Done.\n')
%% open file
fprintf('Open file mpii_human_pose_annotations.json\n')

farrajota

model.modules[1].parameters = function() return nil end -- freezes the layer when using optim 
model.modules[1].accGradParameters = function() end -- overwrite this to reduce computations