def infer_parser() -> argparse.ArgumentParser:
parser = config.default_parser()
parser.add_argument(
"-s", "--source_path",
help="Target to infer",
default="test_examples/images"
)
parser.add_argument(
"-t", "--target_path",
help="Target path to save anonymized result.\
Defaults to subdirectory of config file."
)
parser.add_argument(
"--step", default=None, type=int,
help="Set validation checkpoint to load. Defaults to most recent"
)
return parser
new_sd[key.replace("encoder.", "").replace(".conv.layers", "")] = value
continue
for subkey, new_subkey in mapping.items():
if subkey in key:
old_key = key
key = key.replace(subkey, new_subkey)
break
if "decoder.to_rgb" in key:
continue
new_sd[key] = value
return super().load_state_dict(new_sd, strict=strict)
if __name__ == "__main__":
from deep_privacy.config import Config, default_parser
args = default_parser().parse_args()
cfg = Config.fromfile(args.config_path)
g = MSGGenerator(cfg).cuda()
g.extend()
g.cuda()
imsize = g.current_imsize
batch = dict(
mask=torch.ones((8, 1, imsize, imsize)).cuda(),
condition=torch.randn((8, 3, imsize, imsize)).cuda(),
landmarks=torch.randn((8, 14)).cuda()
)
print(g(**batch).shape)
target_paths.append(target_path.joinpath(impath.name))
else:
target_paths = [pathlib.Path(target_path)]
return image_paths, mask_paths, target_paths
def is_same_shape(images: typing.List[np.ndarray]):
shape1 = images[0].shape
for im in images:
if im.shape != shape1:
return False
return True
if __name__ == "__main__":
parser = default_parser()
parser.add_argument(
"-i",
"--image_path",
default="data/validation_datasets/celebA-HQ",
)
parser.add_argument(
"-m",
"--mask_path",
default=None,
help=
"Path to mask dir/file. Sets the default to _mask file or image_path/../mask"
)
parser.add_argument("-t", "--target_path", default=None)
parser.add_argument(
"--step",
import torch
from deep_privacy import config
from deep_privacy.engine import Trainer, ProgressiveTrainer
# Debug
parser = config.default_parser()
parser.add_argument("--debug", default=False, action="store_true")
args = parser.parse_args()
if args.debug:
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
torch.set_printoptions(precision=10)
cfg = config.Config.fromfile(args.config_path)
cfg.dump()
if cfg.trainer.progressive.enabled:
trainer = ProgressiveTrainer(cfg)
else:
trainer = Trainer(cfg)
trainer.train()