synthesize.py

import os
import time
import argparse
import tensorflow as tf
from src.FourierCPPN import FourierCPPN
from src.CPPN import CPPN
from src.Dataset import Dataset


# COMMAND LINE ARGS
parser = argparse.ArgumentParser(description='training')
parser.add_argument('-g', '--gpu', default=1, type=int)
parser.add_argument('-b', '--batch_size', default=1, type=int)
parser.add_argument('-i', '--iterations', default=100000, type=int)
parser.add_argument('-lr', '--learning_rate', default=5e-4, type=float)
parser.add_argument('-logf', '--log_frequency', default=10, type=int)
parser.add_argument('-printf', '--print_frequency', default=10, type=int)
parser.add_argument('-snapf', '--snapshot_frequency', default=2000, type=int)
parser.add_argument('-writef', '--write_frequency', default=100, type=int)
parser.add_argument('-log_dir', '--log_dir', default='logs', type=str)
parser.add_argument('-snap_dir', '--snapshot_dir',
                    default='snapshots', type=str)
parser.add_argument('-data_dir', '--data_dir', default='data', type=str)
parser.add_argument('-dataset_dir', '--dataset_dir', default='dataset',
                    type=str)
parser.add_argument('-id', '--run_id', default=time.strftime('%d%b-%X'),
                    type=str)
parser.add_argument('-train', '--train', default=True, type=bool)
parser.add_argument('-cppn', '--use_cppn', default=False, type=bool)
parser.add_argument('-bfgs', '--use_bfgs', default=True, type=bool)
parser.add_argument('--cppn_input_dimensions', default='225,225', type=str)
parser.add_argument('--cppn_input_coordinate_range', default='112,112',
                    type=str)
parser.add_argument('--cppn_coord_scale_factor', default='(3**0.5)/112',
                    type=str)
parser.add_argument('--cppn_num_layers', default=8, type=int)
parser.add_argument('--cppn_num_neurons', default=24, type=int)
parser.add_argument('--cppn_activation', default='atan_concat', type=str)
parser.add_argument('--cppn_latent_size', default=2, type=int)
parser.add_argument('--style_layers',
                    default='conv1_1/Relu,pool1,pool2,pool3,pool4',
                    type=str)
parser.add_argument('--content_layers',
                    default='conv1_1/Relu,pool1,pool2,pool3,pool4',
                    type=str)
parser.add_argument('--cppn_fourier_dimensions', default='10,10', type=str)
parser.add_argument('--cppn_fourier_coordinate_range', default='-5,4',
                    type=str)

args = parser.parse_args()

# USER SETTINGS
my_config = {}
my_config['batch_size'] = args.batch_size
my_config['learning_rate'] = args.learning_rate
my_config['write_frequency'] = args.write_frequency
my_config['snapshot_frequency'] = args.snapshot_frequency
my_config['print_frequency'] = args.print_frequency
my_config['log_frequency'] = args.log_frequency
my_config['run_id'] = args.run_id
my_config['iterations'] = args.iterations
my_config['log_dir'] = args.log_dir
my_config['snap_dir'] = args.snapshot_dir
my_config['data_dir'] = args.data_dir
my_config['dataset_dir'] = args.dataset_dir
my_config['train'] = args.train
my_config['use_bfgs'] = args.use_bfgs
my_config['cppn_input_dimensions'] = args.cppn_input_dimensions
my_config['cppn_input_coordinate_range'] = args.cppn_input_coordinate_range
my_config['cppn_coord_scale_factor'] = args.cppn_coord_scale_factor
my_config['cppn_num_layers'] = args.cppn_num_layers
my_config['cppn_num_neurons'] = args.cppn_num_neurons
my_config['cppn_activation'] = args.cppn_activation
my_config['cppn_latent_size'] = args.cppn_latent_size
my_config['style_layers'] = args.style_layers
my_config['content_layers'] = args.content_layers
my_config['cppn_fourier_dimensions'] = args.cppn_fourier_dimensions
my_config['cppn_fourier_coordinate_range'] = args.cppn_fourier_coordinate_range

# GPU SETTINGS
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
tf_config.allow_soft_placement = True
os.environ['CUDA_DEVICE_ORDER'] = 'PCI_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu)

# DATASET SETUP
dataset_path = os.path.join(my_config['data_dir'],
                            my_config['dataset_dir'])
dataset = Dataset(training_path=dataset_path, config=my_config)

# BUILD GRAPH
if args.use_cppn:
    cppn = CPPN(dataset=dataset,
                tf_config=tf_config,
                my_config=my_config)
else:
    cppn = FourierCPPN(dataset=dataset,
                       tf_config=tf_config,
                       my_config=my_config)

if not args.train:
    # PREDICT
    cppn.predict(os.path.join(my_config['snap_dir'], '846577'))
else:
    # NOTE KEEPING
    notes_path = os.path.join(my_config['log_dir'],
                              my_config['run_id'] + '.txt')
    with open(notes_path, 'w') as fp:
        notes = \
            '\n'.join('{!s}={!r}'.format(key, val) for (key, val)
                      in my_config.items())
        fp.write(notes)
        fp.write('\n')

    # TRAIN
    cppn.train()