def main(args):
print(args)
cudnn.benchmark = True
torch.manual_seed(args.seed)
if args.mode == "train":
args.img_datatype = find_img_datatype(args.train_img_dir)
elif args.mode == "sample":
args.img_datatype = find_img_datatype(args.src_dir)
solver = Solver(args)
if args.mode == 'train':
assert len(subdirs(args.train_img_dir)) == args.num_domains
assert len(subdirs(args.val_img_dir)) == args.num_domains
loaders = Munch(src=get_train_loader(root=args.train_img_dir,
which='source',
img_type=args.img_datatype,
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
ref=get_train_loader(root=args.train_img_dir,
which='reference',
img_type=args.img_datatype,
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
val=get_test_loader(root=args.val_img_dir,
img_type=args.img_datatype,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers))
solver.train(loaders)
elif args.mode == 'sample':
assert len(subdirs(args.src_dir)) == args.num_domains
assert len(subdirs(args.ref_dir)) == args.num_domains
loaders = Munch(src=get_test_loader(root=args.src_dir,
img_type=args.img_datatype,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers),
ref=get_test_loader(root=args.ref_dir,
img_type=args.img_datatype,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers))
solver.sample(loaders)
elif args.mode == 'eval':
solver.evaluate()
elif args.mode == 'align':
from core.wing import align_faces
align_faces(args, args.inp_dir, args.out_dir)
else:
raise NotImplementedError
def main(_):
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.4)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
if cfg.env_type == 'simple':
env = SimpleGymEnvironment(cfg)
else:
env = GymEnvironment(cfg)
if not os.path.exists('/tmp/model_dir'):
os.mkdir('/tmp/model_dir')
solver = Solver(cfg, env, sess, '/tmp/model_dir')
solver.train()
def main(args):
print(args)
cudnn.benchmark = True
torch.manual_seed(args.seed)
solver = Solver(args)
if args.mode == 'train':
sintel_path = "/home/tomstrident/datasets/"
video_id = "temple_2"
test_loader = getTestDatasetLoader(sintel_path, video_id)
train_loader, eval_loader = get_loaderFC2(
args.data_dir, args.style_dir, args.temp_dir, args.batch_size,
args.num_workers, args.num_domains, args.mode)
print("start training ...")
print("args.num_domains:", args.num_domains)
solver.train([train_loader, test_loader])
elif args.mode == 'sample':
assert len(subdirs(args.src_dir)) == args.num_domains
assert len(subdirs(args.ref_dir)) == args.num_domains
loaders = Munch(src=get_test_loader(root=args.src_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers),
ref=get_test_loader(root=args.ref_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers))
solver.sample(loaders)
elif args.mode == 'eval':
_, eval_loader = get_loaderFC2(args.data_dir, args.style_dir,
args.temp_dir, args.batch_size,
args.num_workers, args.num_domains,
args.mode)
print("len(eval_loader)", len(eval_loader))
solver.evaluate(loader=eval_loader)
#solver.eval_sintel()
elif args.mode == 'align':
from core.wing import align_faces
align_faces(args, args.inp_dir, args.out_dir)
else:
raise NotImplementedError
def main(args):
print(args)
cudnn.benchmark = True
torch.manual_seed(args.seed)
solver = Solver(args)
if args.mode == 'train':
loaders = Munch(src=get_train_loader(root=args.train_img_dir,
which='source',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers,
dataset_dir=args.dataset_dir),
ref=get_train_loader(root=args.train_img_dir,
which='reference',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers,
dataset_dir=args.dataset_dir),
val=get_val_loader(root=args.val_img_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers,
dataset_dir=args.dataset_dir))
solver.train(loaders)
elif args.mode == 'sample': ## 'styling_ref' 모드에 맞게 hyun 추가
solver.sample()
# hyun 추가
#parsing(respth='./results/label/src', dspth= os.path.join(args.src_dir,trg_domain)) # parsing src_image
#parsing(respth='./results/label/others', dspth= os.path.join(args.result_dir,trg_domain) # parsing fake_image
#reconstruct() # 'styling' 모드
elif args.mode == 'eval':
fid_values, fid_mean = solver.evaluate()
for key, value in fid_values.items():
print(key, value)
else:
raise NotImplementedError
def train_net(cfg):
net = ReconstructionNet(cfg)
print('Network definition:')
print(inspect.getsource(ReconstructionNet.network_definition))
# Generate the solver
solver = Solver(cfg, net)
# Prefetching data processes
#
# Create worker and data queue for data processing. For training data, use
# multiple processes to speed up the loading. For validation data, use 1
# since the queue will be popped every TRAIN.NUM_VALIDATION_ITERATIONS.
global train_queue, val_queue, train_processes, val_processes
train_queue = Queue(cfg.TRAIN.QUEUE_SIZE)
val_queue = Queue(cfg.TRAIN.QUEUE_SIZE)
train_processes = make_processes(cfg,
train_queue,
category_model_id_pair(
cfg.DIR.DATASET_TAXONOMY_FILE_PATH,
cfg.DIR.DATASET_QUERY_PATH,
cfg.TRAIN.DATASET_PORTION),
cfg.TRAIN.NUM_WORKER,
repeat=True)
val_processes = make_processes(cfg,
val_queue,
category_model_id_pair(
cfg.DIR.DATASET_TAXONOMY_FILE_PATH,
cfg.DIR.DATASET_QUERY_PATH,
cfg.TEST.DATASET_PORTION),
1,
repeat=True,
train=False)
# Train the network
solver.train(train_queue, val_queue)
# Cleanup the processes and the queue.
kill_processes(train_queue, train_processes)
kill_processes(val_queue, val_processes)
def main(config):
# For fast training.
cudnn.benchmark = True
# Create directories if not exist.
if not os.path.exists(config.log_dir):
os.makedirs(config.log_dir)
if not os.path.exists(config.model_save_dir):
os.makedirs(config.model_save_dir)
if not os.path.exists(config.sample_dir):
os.makedirs(config.sample_dir)
if not os.path.exists(config.result_dir):
os.makedirs(config.result_dir)
# Data loader.
celeba_loader = None
rafd_loader = None
if config.dataset in ['CelebA', 'Both']:
celeba_loader = get_loader(config.celeba_image_dir, config.attr_path, config.selected_attrs,
config.celeba_crop_size, config.image_size, config.batch_size,
'CelebA', config.mode, config.num_workers)
if config.dataset in ['RaFD', 'Both']:
rafd_loader = get_loader(config.rafd_image_dir, None, None,
config.rafd_crop_size, config.image_size, config.batch_size,
'RaFD', config.mode, config.num_workers)
# Solver for training and testing StarGAN.
solver = Solver(celeba_loader, rafd_loader, config)
if config.mode == 'train':
if config.dataset in ['CelebA', 'RaFD']:
solver.train()
elif config.dataset in ['Both']:
solver.train_multi()
elif config.mode == 'test':
if config.dataset in ['CelebA', 'RaFD']:
solver.test()
elif config.dataset in ['Both']:
solver.test_multi()
def main(args):
print(args)
cudnn.benchmark = True
if args.mode == 'train':
torch.manual_seed(args.seed)
solver = Solver(args)
transform = transforms.Compose([
transforms.Resize([args.img_size, args.img_size]),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])
if args.mode == 'train':
assert len(subdirs(args.train_img_dir)) == args.num_domains
assert len(subdirs(args.val_img_dir)) == args.num_domains
if args.resume_iter > 0:
solver._load_checkpoint(args.resume_iter)
loaders = Munch(src=get_train_loader(root=args.train_img_dir,
which='source',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
ref=get_train_loader(root=args.train_img_dir,
which='reference',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
val=get_test_loader(root=args.val_img_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers))
solver.train(loaders)
elif args.mode == 'eval':
solver.evaluate()
elif args.mode == 'align':
from core.wing import align_faces
align_faces(args, args.inp_dir, args.out_dir)
elif args.mode == 'inter': # interpolation
save_dir = args.save_dir
if not os.path.exists(save_dir):
os.mkdir(save_dir)
solver._load_checkpoint(args.resume_iter)
nets_ema = solver.nets_ema
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
image_name = os.path.basename(args.input)
image = Variable(
transform(Image.open(
args.input).convert('RGB')).unsqueeze(0).to(device))
masks = nets_ema.fan.get_heatmap(image) if args.w_hpf > 0 else None
y1 = torch.tensor([args.y1]).long().cuda()
y2 = torch.tensor([args.y2]).long().cuda()
outputs = interpolations(nets_ema,
args.latent_dim,
image,
masks,
lerp_step=0.1,
y1=y1,
y2=y2,
lerp_mode=args.lerp_mode)
path = os.path.join(save_dir, image_name)
vutils.save_image(outputs.data, path, padding=0)
elif args.mode == 'test':
save_dir = args.save_dir
if not os.path.exists(save_dir):
os.mkdir(save_dir)
solver._load_checkpoint(args.resume_iter)
nets_ema = solver.nets_ema
image_name = os.path.basename(args.input)
image = Variable(
transform(Image.open(
args.input).convert('RGB')).unsqueeze(0)).cuda()
masks = nets_ema.fan.get_heatmap(image) if args.w_hpf > 0 else None
image_ref = None
if args.test_mode == 'reference':
image_ref = Variable(
transform(Image.open(
args.input_ref).convert("RGB")).unsqueeze(0)).cuda()
fake = test_single(nets_ema, image, masks, args.latent_dim, image_ref,
args.target_domain, args.single_mode)
fake = torch.clamp(fake * 0.5 + 0.5, 0, 1)
path = os.path.join(save_dir, image_name)
vutils.save_image(fake.data, path, padding=0)
elif args.mode == 'video':
save_dir = args.save_dir
if not os.path.exists(save_dir):
os.mkdir(save_dir)
solver._load_checkpoint(args.resume_iter)
nets_ema = solver.nets_ema
image_name = os.path.basename(args.input)
image = Variable(
transform(Image.open(
args.input).convert('RGB')).unsqueeze(0)).cuda()
masks = nets_ema.fan.get_heatmap(image) if args.w_hpf > 0 else None
y1 = torch.tensor([args.y1]).long().cuda()
y2 = torch.tensor([args.y2]).long().cuda()
outputs = interpolations_loop(nets_ema,
args.latent_dim,
image,
masks,
lerp_step=0.02,
y1=y1,
y2=y2,
lerp_mode=args.lerp_mode)
outputs = torch.cat(outputs)
outputs = tensor2ndarray255(outputs)
path = os.path.join(save_dir, '{}-video.mp4'.format(image_name))
save_video(path, outputs)
else:
raise NotImplementedError
def main(args):
print(args)
cudnn.benchmark = True
torch.manual_seed(args.seed)
solver = Solver(args)
#create csv file
with open(args.loss_csv_path, 'wb') as csvfile:
filewriter = csv.writer(csvfile,
delimiter=',',
quotechar='|',
quoting=csv.QUOTE_MINIMAL)
with open(args.loss_csv_path, 'a') as file:
writer = csv.writer(file)
writer.writerow([
"epoch", "d_loss_z_trg", "d_loss_x_ref", "g_loss_z_trg",
"g_loss_x_ref"
])
if args.mode == 'train':
assert len(subdirs(args.train_img_dir)) == args.num_domains
assert len(subdirs(args.val_img_dir)) == args.num_domains
loaders = Munch(src=get_train_loader(root=args.train_img_dir,
which='source',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
src_skt=get_train_loader(
root=args.train_sketch_img_dir,
which='source',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
ref=get_train_loader(root=args.train_img_dir,
which='reference',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
val=get_test_loader(root=args.val_img_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers))
solver.train(loaders)
elif args.mode == 'sample':
assert len(subdirs(args.src_dir)) == args.num_domains
assert len(subdirs(args.ref_dir)) == args.num_domains
loaders = Munch(src=get_test_loader(root=args.src_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers),
ref=get_test_loader(root=args.ref_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers))
solver.sample(loaders)
elif args.mode == 'eval':
solver.evaluate()
elif args.mode == 'align':
from core.wing import align_faces
align_faces(args, args.inp_dir, args.out_dir)
else:
raise NotImplementedError
def main(args):
print(args)
#wandb.init(project="stargan", entity="stacey", config=args, name=args.model_name)
#cfg = wandb.config
#cfg.update({"dataset" : "afhq", "type" : "train"})
cudnn.benchmark = True
torch.manual_seed(args.seed)
solver = Solver(args)
if args.mode == 'train':
assert len(subdirs(args.train_img_dir)) == args.num_domains
assert len(subdirs(args.val_img_dir)) == args.num_domains
loaders = Munch(src=get_train_loader(root=args.train_img_dir,
which='source',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
ref=get_train_loader(root=args.train_img_dir,
which='reference',
img_size=args.img_size,
batch_size=args.batch_size,
prob=args.randcrop_prob,
num_workers=args.num_workers),
val=get_test_loader(root=args.val_img_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=True,
num_workers=args.num_workers))
solver.train(loaders)
elif args.mode == 'sample':
assert len(subdirs(args.src_dir)) == args.num_domains
assert len(subdirs(args.ref_dir)) == args.num_domains
loaders = Munch(src=get_test_loader(root=args.src_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers),
ref=get_test_loader(root=args.ref_dir,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers))
solver.sample(loaders)
elif args.mode == 'eval':
solver.evaluate(args)
elif args.mode == 'align':
from core.wing import align_faces
align_faces(args, args.inp_dir, args.out_dir)
elif args.mode == 'custom':
# override some default arguments
wandb.init(project="stargan", config=args, name=args.model_name)
# src or ref may each be a dir or an image
# make temporary folders for images
if os.path.isfile(args.custom_src):
src_dir = "tmp_src"
full_src = src_dir + "/src"
if os.path.exists(src_dir):
shutil.rmtree(src_dir)
os.makedirs(full_src)
shutil.copy2(args.custom_src, full_src)
src_images = src_dir
else:
src_images = args.custom_src
if os.path.isfile(args.custom_ref):
ref_dir = "tmp_ref"
full_ref = ref_dir + "/ref"
if os.path.exists(ref_dir):
shutil.rmtree(ref_dir)
os.makedirs(full_ref)
shutil.copy2(args.custom_ref, full_ref)
if args.extend_domain:
# make some extra domains
for d in [ref_dir + "/ref2", ref_dir + "/ref3"]:
os.makedirs(d)
shutil.copy2(args.custom_ref, d)
ref_images = ref_dir
else:
ref_images = args.custom_ref
loaders = Munch(src=get_test_loader(root=src_images,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers),
ref=get_test_loader(root=ref_images,
img_size=args.img_size,
batch_size=args.val_batch_size,
shuffle=False,
num_workers=args.num_workers))
solver.custom(loaders)
else:
raise NotImplementedError
