def main():
args = get_args()
save_args(args, "evaluate")
evaluate(args)
def main():
args = get_args()
save_args(args, "train")
train(args)
def main():
# Args
args = get_args()
save_args(args)
# Context
ctx = extension_context(args.context,
device_id=args.device_id,
type_config=args.type_config)
nn.set_default_context(ctx)
nn.set_auto_forward(True)
# Data Itrator
di = data_iterator(args.img_path,
args.batch_size,
imsize=(args.imsize, args.imsize),
num_samples=args.train_samples,
dataset_name=args.dataset_name)
# Model
generator = Generator(use_bn=args.use_bn,
last_act=args.last_act,
use_wscale=args.not_use_wscale,
use_he_backward=args.use_he_backward)
discriminator = Discriminator(use_ln=args.use_ln,
alpha=args.leaky_alpha,
use_wscale=args.not_use_wscale,
use_he_backward=args.use_he_backward)
# Solver
solver_gen = S.Adam(alpha=args.learning_rate,
beta1=args.beta1,
beta2=args.beta2)
solver_dis = S.Adam(alpha=args.learning_rate,
beta1=args.beta1,
beta2=args.beta2)
# Monitor
monitor = Monitor(args.monitor_path)
monitor_loss_gen = MonitorSeries("Generator Loss", monitor, interval=10)
monitor_loss_dis = MonitorSeries("Discriminator Loss",
monitor,
interval=10)
monitor_p_fake = MonitorSeries("Fake Probability", monitor, interval=10)
monitor_p_real = MonitorSeries("Real Probability", monitor, interval=10)
monitor_time = MonitorTimeElapsed("Training Time per Resolution",
monitor,
interval=1)
monitor_image_tile = MonitorImageTileWithName("Image Tile",
monitor,
num_images=4,
normalize_method=lambda x:
(x + 1.) / 2.)
# TODO: use argment
resolution_list = [4, 8, 16, 32, 64, 128]
channel_list = [512, 512, 256, 128, 64, 32]
trainer = Trainer(di,
generator,
discriminator,
solver_gen,
solver_dis,
args.monitor_path,
monitor_loss_gen,
monitor_loss_dis,
monitor_p_fake,
monitor_p_real,
monitor_time,
monitor_image_tile,
resolution_list,
channel_list,
n_latent=args.latent,
n_critic=args.critic,
save_image_interval=args.save_image_interval,
hyper_sphere=args.hyper_sphere,
l2_fake_weight=args.l2_fake_weight)
# TODO: use images per resolution?
trainer.train(args.epoch_per_resolution)
def main():
args = get_args()
save_args(args, "generate")
generate(args)
best_ve = ve
# Training forward
image.d, label.d = data.next()
solver.zero_grad()
loss.forward(clear_no_need_grad=True)
loss.backward(clear_buffer=True)
solver.weight_decay(args.weight_decay)
solver.update()
e = categorical_error(pred.d, label.d)
monitor_loss.add(i, loss.d.copy())
monitor_err.add(i, e)
monitor_time.add(i)
ve = 0.0
for j in range(int(n_valid / args.batch_size)):
vimage.d, vlabel.d = vdata.next()
vpred.forward(clear_buffer=True)
ve += categorical_error(vpred.d, vlabel.d)
ve /= int(n_valid / args.batch_size)
monitor_verr.add(i, ve)
parameter_file = os.path.join(args.model_save_path,
'params_{:06}.h5'.format(args.max_iter))
nn.save_parameters(parameter_file)
if __name__ == '__main__':
args = get_args()
save_args(args, "train")
train(args)
def main():
args = get_args()
save_args(args, "morph")
morph(args)
def main():
args = get_args()
save_args(args, "match")
match(args)
def main():
args = get_args()
save_args(args, "generate")
interpolate(args)
def main():
args = get_args()
save_args(args)
test(args)
