max_dimsize=512,
num_iters=100,
content_weight=0.0,
init_img_type='custom',
)
default_config = StyleConfig()
default_config.update(**custom_params)
styler = Styler(default_config, device)
def zoom(image, zoom_pixels):
# PIL image -> remove borders -> resize -> PIL Image
width, height = image.size
left, right = zoom_pixels, width - zoom_pixels
top, bottom = zoom_pixels, height - zoom_pixels
return image.crop((left, top, right, bottom)).resize((width, height))
content_image = Image.open(util.content_img_path(CONTENT_IMG_NAME))
style_image = Image.open(util.style_img_path(STYLE_IMG_NAME))
last_frame_result = content_image
for frame in range(NUM_FRAMES):
image = styler.style_multiscale(content_image,
style_image,
octaves=NUM_OCTAVES,
init_img=last_frame_result)
last_frame_result = zoom(image, 10)
util.imsave(image, name=f"{OUTPUT_NAME}_f{frame}")
import util
import torch
from PIL import Image
# set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
CONTENT_IMG_NAME = "lion.jpg"
STYLE_IMG_NAME = "starry-night.jpg"
OUTPUT_NAME = "test"
NUM_OCTAVES = 3
params = {
"max_dimsize": 512,
}
default_config = StyleConfig()
default_config.update(**params)
styler = Styler(default_config, device)
content_image = Image.open(util.content_img_path(CONTENT_IMG_NAME))
style_image = Image.open(util.style_img_path(STYLE_IMG_NAME))
output = styler.style_multiscale(content_image,
style_image,
octaves=NUM_OCTAVES,
save_intermediate=True)
util.imsave(output, name=OUTPUT_NAME)