main.lua

local dataset = require 'dataset'
local network = require 'network'
local training = require 'training_optim'
local testing = require 'testing'
local matching = require 'matching'

local pathx = require 'pl.path'
require 'torch'

print('Reading training dataset.')
local train_dataset
local val_dataset
local test_dataset
if pathx.exists('train_dataset.bin') and pathx.exists('val_dataset.bin') then
  print('Using existing train and validation datasets.')
  train_dataset = torch.load('train_dataset.bin')
  val_dataset = torch.load('val_dataset.bin')
else
  train_dataset, val_dataset = dataset.get_dataset(false, true, false)
  torch.save('train_dataset.bin', train_dataset)
  torch.save('val_dataset.bin', val_dataset)
end
print('Using ' .. train_dataset.nbr_elements .. ' training samples.')
print('Using ' .. val_dataset.nbr_elements .. ' validation samples.')

print('Reading testing dataset.')
local test_dataset
if pathx.exists('test_dataset.bin') then
  print('Using existing test dataset.')
  test_dataset = torch.load('test_dataset.bin')
else
  test_dataset = dataset.get_dataset(true, false, false)
  torch.save('test_dataset.bin', test_dataset)
end
print('Using ' .. test_dataset.nbr_elements .. ' testing samples.')


local cnn
if pathx.exists('model.bin') then
  print('Using pretrained network.')
  cnn = torch.load('model.bin')
else
  cnn = network.get_network_multiscale()
  print('Training network.')
  training.train_network(cnn, train_dataset, val_dataset)

  print('Saving network.')
  torch.save('model.bin', cnn)
end


print('Testing network.')
testing.test_network(cnn, test_dataset)
matching.test_matching(cnn, 4, test_dataset)