def __init__(self,
sess,
model_dir="./cgan_model",
max_objs=3,
result_dim=256,
font="ebgaramond.ttf",
font_size=40):
self.sess = sess
self.image_finder = image_finder.ImageFinder()
self.db = MongoClient().albart
self.cgan_model_dir = model_dir
self.cgan = cgan.CGAN(self.sess, n_batches=150)
self.cgan.init_model()
self.lyrics_api = genius_api.LyricsApi()
self.max_objs = max_objs
self.result_dim = result_dim
self.font_size = font_size
self.font_string = font
self.font = ImageFont.truetype(font, font_size)
self.allowed_genres = [
"country", "pop", "rap", "reggae", "indie rock", "hip hop"
]
def main(to_reload=None):
cgan_train = dataset.load_data(TRAIN_FILES['A'],
TRAIN_FILES['B'],
skiprows=1,
transpose=True)
if to_reload: # restore, get fake and cycle outputs
m = cgan.CGAN(meta_graph=to_reload, config_name=CONFIG_FILE_NAME)
print("Loaded!", flush=True)
### use training data as testing data
test_a, test_b = cgan_train.data
fake_a = m.generate_fake_a(test_b)
print('generated fake a dimension: {}'.format(fake_a.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'fake_a.txt',
X=fake_a.T,
delimiter='\t',
comments='')
fake_b = m.generate_fake_b(test_a)
print('generated fake b dimension: {}'.format(fake_b.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'fake_b.txt',
X=fake_b.T,
delimiter='\t',
comments='')
cycle_a = m.generate_cycle_a(test_a)
print('generated cycle a dimension: {}'.format(cycle_a.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'cycle_a.txt',
X=cycle_a.T,
delimiter='\t',
comments='')
cycle_b = m.generate_cycle_b(test_b)
print('generated cycle b dimension: {}'.format(cycle_b.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'cycle_b.txt',
X=cycle_b.T,
delimiter='\t',
comments='')
a2b_final_weights = m.a2b_final_w()
print('a2b generator output weights size: {}'.format(
a2b_final_weights.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'a2b_output_weights.txt',
X=a2b_final_weights.T,
delimiter='\t',
comments='')
b2a_final_weights = m.b2a_final_w()
print('b2a generator output weights size: {}'.format(
b2a_final_weights.shape))
np.savetxt(fname=MODEL_NAME + '_' + 'b2a_output_weights.txt',
X=b2a_final_weights.T,
delimiter='\t',
comments='')
else: # train
"""to try cont'd training, load data from previously saved meta graph"""
m = cgan.CGAN(ARCHITECTURE,
HYPERPARAMS,
config_name=CONFIG_FILE_NAME + '_' + MODEL_NAME,
log_dir=LOG_DIR)
m.train(cgan_train,
max_iter=MAX_ITER,
max_epochs=MAX_EPOCHS,
verbose=True,
save=True,
outdir=METAGRAPH_DIR)
print("Trained!", flush=True)
def main():
start_time = time.time() # clocking start
# mnist data loading
mnist = DataSet(dataset_name='mnist')
mnist = mnist.mnist
# GPU configure
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as s:
# GAN Model
model = cgan.CGAN(s)
# initializing
s.run(tf.global_variables_initializer())
sample_x, _ = mnist.train.next_batch(model.sample_num)
sample_y = np.zeros(shape=[model.sample_num, model.n_classes])
sample_y[:, 3] = 1 # specify label number what u wanna get
sample_z = np.random.uniform(
-1., 1., [model.sample_num, model.z_dim]).astype(np.float32)
d_overpowered = False
for step in range(paras['global_step']):
batch_x, batch_y = mnist.train.next_batch(model.batch_size)
batch_z = np.random.uniform(-1.,
1.,
size=[model.batch_size,
model.z_dim]).astype(np.float32)
# update D network
if not d_overpowered:
s.run(model.d_op,
feed_dict={
model.x: batch_x,
model.c: batch_y,
model.z: batch_z
})
# update G network
s.run(model.g_op, feed_dict={model.c: batch_y, model.z: batch_z})
if step % paras['logging_interval'] == 0:
batch_x, batch_y = mnist.test.next_batch(model.batch_size)
batch_z = np.random.uniform(
-1., 1.,
[model.batch_size, model.z_dim]).astype(np.float32)
d_loss, g_loss, summary = s.run(
[model.d_loss, model.g_loss, model.merged],
feed_dict={
model.x: batch_x,
model.c: batch_y,
model.z: batch_z
})
# update d_overpowered
d_overpowered = d_loss < g_loss / 3
# print loss
print("[+] Step %08d => " % (step),
"D loss : {:.8f}".format(d_loss),
" G loss : {:.8f}".format(g_loss))
# training G model with sample image and noise
samples = s.run(model.G,
feed_dict={
model.c: sample_y,
model.z: sample_z
})
# summary saver
model.writer.add_summary(summary, step)
# export image generated by model G
sample_image_height = model.sample_size
sample_image_width = model.sample_size
sample_dir = dirs['sample_output'] + 'train_{:08d}.png'.format(
step)
# Generated image save
iu.save_images(samples,
size=[sample_image_height, sample_image_width],
image_path=sample_dir)
# model save
model.saver.save(s, dirs['model'], global_step=step)
end_time = time.time() - start_time
# elapsed time
print("[+] Elapsed time {:.8f}s".format(end_time))
# close tf.Session
s.close()
return self.sess.run(self.model.latent,
feed_dict={
self.model.orig: self.X,
self.model.EPS_BS: 1,
self.model.TRAIN: False
})
def get_zdim(self):
return self.model.lat_size
# Step 1: Create theano functions
# Initialize model
# model = IAN(config_path = 'IAN_simple.py', dnn = True)
net = cgan.CGAN()
model = fooler(model=net, log_dir='results/trial/LOGS')
# z_trial = np.eye(100)[45].reshape(-1, 100)
# out = model.sample_at(z_trial)
# imsave('trial.png', out[0])
# imsave('trial.png',out.reshape(5,5,64,64,3).swapaxes(1,2).reshape(5*64,-1,3))
# z_trial = np.random.uniform(-1.,1., size=(25,100))
# out = model.sample_at(z_trial)
# print(out.shape)
# imsave('trial2.png',out.reshape(5,5,64,64,3).swapaxes(1,2).reshape(5*64,-1,3))
# import sys
# sys.exit()
# # Prepare GUI functions
import cgan
import sys
import network
from data_interface import mnist
DATA_SET_SIZE = 65000
if __name__ == '__main__':
tf.reset_default_graph()
# load MNIST
train_validation_set = mnist.train('../../dataset/MNIST_Dataset/train/')
test_set = mnist.test('../../dataset/MNIST_Dataset/test/')
data_set = train_validation_set.concatenate(test_set)
# run cycle for current parameter settings
gan = cgan.CGAN(generator_fn=network.cgan_generator,
discriminator_fn=network.cgan_discriminator)
train_params = {
'lr': 0.0002,
'beta1': 0.5,
'batch_size': 100,
'max_epoch': 100,
}
g_model_params = {
'input_shape': ((200, ), (10, )),
'output_shape': (784, ),
'output_image_shape': (28, 28, 1),
'summ_shape_per_class': (10, 1),
'z_dense_width': 256,
'y_dense_width': 256,
'mid_dense_widths': [512, 1024],
}