# cnn2_output = load_RGB('data/kosode/cnn2_after_train_norm', file_num * motif_num)
# cnn1_W = loadW('data/kosode_motif2/train/cnn1_after_train')
# cnn2_W = loadW('data/kosode_motif2/train/cnn2_after_train')
makeFolder()
# 時間計測
time1 = time.clock()
print '~~~CNN1~~~'
cnn1 = CNN(train_set, filter_shape, filter_shift_list[0], input_shape, node_shape[1], cnn_pre_train_lr, cnn_pre_train_epoch, isRGB)
output_list = cnn1.output()
cnn_saveColorImage(output_list, node_shape[1], 'cnn1_before_train')
output_list_norm = local_contrast_normalization(output_list)
cnn_saveColorImage(output_list_norm, node_shape[1], 'cnn1_before_training_norm')
cnn1.pre_train()
# cnn1.setW(cnn1_W)
output_list = cnn1.output()
cnn_saveColorImage(output_list, node_shape[1], 'cnn1_after_train')
output_list_norm = local_contrast_normalization(output_list)
cnn_saveColorImage(output_list_norm, node_shape[1], 'cnn1_after_train_norm')
print '~~~CNN2~~~'
cnn2 = CNN(cnn1.output(), filter_shape, filter_shift_list[1], node_shape[1], node_shape[2], cnn_pre_train_lr, cnn_pre_train_epoch, isRGB)
# node_shape = ((80,52), (74,46), (35,21))
node_shape = ((80, 52), (74, 46), (34, 20))
filter_shift_list = ((1, 1), (2, 2))
input_shape = [80, 52]
filter_shape = [7, 7]
data_list = load_image(data_path, file_num, isRGB)
makeFolder()
# 時間計測
time1 = time.clock()
cnn1 = CNN(data_list, filter_shape, filter_shift_list[0], input_shape, node_shape[1], pre_train_lr, pre_train_epoch)
output_list = cnn1.output()
saveImage(output_list, node_shape[1], 'cnn1_before_training')
cnn1.pre_train()
output_list = cnn1.output()
saveImage(output_list, node_shape[1], 'cnn1_after_training')
# for i in xrange(pre_train_epoch):
# cnn1.pre_train()
# output_list = cnn1.output()
# saveImage(output_list, (74,46))
cnn2 = CNN(cnn1.output(), filter_shape, filter_shift_list[1], node_shape[1], node_shape[2], pre_train_lr, pre_train_epoch)
output_list = cnn2.output()
saveImage(output_list, node_shape[2], 'cnn2_before_train')
