def figure_3_d(content_image, style_image):
content_image_yuv = rgb_to_yuv(content_image)
content_luminance = content_image_yuv[:, :1].repeat(1, 3, 1, 1)
content_chromaticity = content_image_yuv[:, 1:]
style_luminance = rgb_to_grayscale(style_image, num_output_channels=3)
print("Replicating Figure 3 (d)")
output_luminance = paper.nst(
content_luminance,
style_luminance,
impl_params=args.impl_params,
)
output_luminance = torch.mean(output_luminance, dim=1, keepdim=True)
output_chromaticity = resize(content_chromaticity,
output_luminance.size()[2:])
output_image_yuv = torch.cat((output_luminance, output_chromaticity),
dim=1)
output_image = yuv_to_rgb(output_image_yuv)
filename = make_output_filename(
"gatys_et_al_2017",
"fig_3",
"d",
impl_params=args.impl_params,
)
output_file = path.join(args.image_results_dir, filename)
save_result(output_image, output_file)
def figure_2_ef(
label,
content_image,
content_building_guide,
content_sky_guide,
style_building_image,
style_building_guide,
style_sky_image,
style_sky_guide,
):
content_guides = {
"building": content_building_guide,
"sky": content_sky_guide
}
style_images_and_guides = {
"building": (style_building_image, style_building_guide),
"sky": (style_sky_image, style_sky_guide),
}
print(f"Replicating Figure 2 ({label})")
output_image = paper.guided_nst(
content_image,
content_guides,
style_images_and_guides,
impl_params=args.impl_params,
)
filename = make_output_filename(
"gatys_et_al_2017",
"fig_2",
label,
impl_params=args.impl_params,
)
output_file = path.join(args.image_results_dir, filename)
save_result(output_image, output_file)
def figure_3(args):
images = paper.images()
images.download(args.image_source_dir)
hyper_parameters = paper.hyper_parameters(args.impl_params)
content_image = images["neckarfront"].read(
size=hyper_parameters.nst.image_size, device=args.device)
style_image = images["composition_vii"].read(
size=hyper_parameters.nst.image_size, device=args.device)
style_layers = hyper_parameters.style_loss.layers
layer_configs = [
style_layers[:idx + 1] for idx in range(len(style_layers))
]
score_weights = (1e5, 1e4, 1e3, 1e2)
params = "implementation" if args.impl_params else "paper"
for layers, score_weight in itertools.product(layer_configs,
score_weights):
row_label = layers[-1]
column_label = f"{1.0 / score_weight:.0e}"
print(
f"Replicating Figure 3 image in row {row_label} and column {column_label} "
f"with {params} parameters.")
hyper_parameters.style_loss.layers = layers
if args.impl_params:
hyper_parameters.style_loss.layer_weights = paper.compute_layer_weights(
layers)
hyper_parameters.style_loss.score_weight = score_weight
output_image = paper.nst(
content_image,
style_image,
impl_params=args.impl_params,
hyper_parameters=hyper_parameters,
)
filename = utils.make_output_filename(
"gatys_ecker_betghe_2016",
"fig_3",
row_label,
column_label,
impl_params=args.impl_params,
)
save_result(
output_image,
path.join(args.image_results_dir, filename),
)
def figure_3_c(content_image, style_image):
print("Replicating Figure 3 (c)")
output_image = paper.nst(
content_image,
style_image,
impl_params=args.impl_params,
)
filename = make_output_filename(
"gatys_et_al_2017",
"fig_3",
"c",
impl_params=args.impl_params,
)
output_file = path.join(args.image_results_dir, filename)
save_result(output_image, output_file)
def figure_3_e(content_image, style_image, method="cholesky"):
style_image = match_channelwise_statistics(style_image, content_image,
method)
print("Replicating Figure 3 (e)")
output_image = paper.nst(
content_image,
style_image,
impl_params=args.impl_params,
)
filename = make_output_filename(
"gatys_et_al_2017",
"fig_3",
"e",
impl_params=args.impl_params,
)
output_file = path.join(args.image_results_dir, filename)
save_result(output_image, output_file)
def figure_2(args):
images = paper.images()
images.download(args.image_source_dir)
hyper_parameters = paper.hyper_parameters()
content_image = images["neckarfront"].read(
size=hyper_parameters.nst.image_size, device=args.device)
class StyleImage:
def __init__(self, label, image, score_weight):
self.label = label
self.image = image.read(size=hyper_parameters.nst.image_size,
device=args.device)
self.score_weight = score_weight
style_images = (
StyleImage("B", images["shipwreck"], 1e3),
StyleImage("C", images["starry_night"], 1e3),
StyleImage("D", images["the_scream"], 1e3),
StyleImage("E", images["femme_nue_assise"], 1e4),
StyleImage("F", images["composition_vii"], 1e4),
)
for style_image in style_images:
print(f"Replicating Figure 2 {style_image.label}")
hyper_parameters.style_loss.score_weight = style_image.score_weight
output_image = paper.nst(
content_image,
style_image.image,
impl_params=args.impl_params,
hyper_parameters=hyper_parameters,
)
filename = utils.make_output_filename(
"gatys_ecker_betghe_2016",
"fig_2",
style_image.label,
impl_params=args.impl_params,
)
save_result(
output_image,
path.join(args.image_results_dir, filename),
)
def figure_6(args):
images = paper.images()
images.download(args.image_source_dir)
positions = ("top", "bottom")
image_pairs = (
(images["blue_bottle"], images["self-portrait"]),
(images["s"], images["composition_viii"]),
)
for position, image_pair in zip(positions, image_pairs):
content_image = image_pair[0].read(device=args.device)
style_image = image_pair[1].read(device=args.device)
print(
f"Replicating the {position} half of figure 6 "
f"with {'implementation' if args.impl_params else 'paper'} parameters"
)
hyper_parameters = paper.hyper_parameters(impl_params=args.impl_params)
if args.impl_params:
# https://github.com/pmeier/CNNMRF/blob/fddcf4d01e2a6ce201059d8bc38597f74a09ba3f/run_trans.lua#L66
hyper_parameters.content_loss.layer = "relu4_2"
hyper_parameters.target_transforms.num_scale_steps = 1
hyper_parameters.target_transforms.num_rotate_steps = 1
output_image = paper.nst(
content_image,
style_image,
impl_params=args.impl_params,
hyper_parameters=hyper_parameters,
)
filename = make_output_filename(
"li_wand_2016",
"fig_6",
position,
impl_params=args.impl_params,
)
output_file = path.join(args.image_results_dir, filename)
print(f"Saving result to {output_file}")
write_image(output_image, output_file)
print("#" * int(os.environ.get("COLUMNS", "80")))
def training(args):
contents = ("karya", "tiger", "neckarfront", "bird", "kitty")
styles = (
"candy",
"the_scream",
"jean_metzinger",
"mosaic",
"pleades",
"starry",
"turner",
)
dataset = paper.dataset(
args.dataset_dir,
impl_params=args.impl_params,
instance_norm=args.instance_norm,
)
image_loader = paper.image_loader(
dataset,
impl_params=args.impl_params,
instance_norm=args.instance_norm,
pin_memory=str(args.device).startswith("cuda"),
)
images = paper.images()
images.download(args.image_source_dir)
for style in styles:
style_image = images[style].read(device=args.device)
transformer = paper.training(
image_loader,
style_image,
impl_params=args.impl_params,
instance_norm=args.instance_norm,
)
model_name = f"ulyanov_et_al_2016__{style}"
if args.impl_params:
model_name += "__impl_params"
if args.instance_norm:
model_name += "__instance_norm"
utils.save_state_dict(transformer, model_name, root=args.model_dir)
for content in contents:
content_image = images[content].read(device=args.device)
output_image = paper.stylization(
content_image,
transformer,
impl_params=args.impl_params,
instance_norm=args.instance_norm,
)
filename = utils.make_output_filename(
"ulyanov_et_al_2016",
style,
content,
impl_params=args.impl_params,
instance_norm=args.instance_norm,
)
output_file = path.join(args.image_results_dir, filename)
image.write_image(output_image, output_file)