content_image = load_image(args.content_image)
style_image = load_image(args.style_image)
# use existing if available
if (load_segmentation):
print("Load segmentation from files.")
content_segmentation_image = cv2.imread(content_segmentation_filename)
style_segmentation_image = cv2.imread(style_segmentation_filename)
content_segmentation_masks = extract_segmentation_masks(content_segmentation_image)
style_segmentation_masks = extract_segmentation_masks(style_segmentation_image)
# otherwise compute it
else:
print("Create segmentation.")
content_segmentation, style_segmentation = compute_segmentation(args.content_image, args.style_image)
cv2.imwrite(change_filename(args.seg_dir, args.content_image, '_seg_raw', '.png'), content_segmentation)
cv2.imwrite(change_filename(args.seg_dir, args.style_image, '_seg_raw', '.png'), style_segmentation)
content_segmentation_masks, style_segmentation_masks = merge_segments(content_segmentation, style_segmentation,
args.semantic_thresh, args.similarity_metric)
cv2.imwrite(change_filename(args.seg_dir, args.content_image, '_seg', '.png'),
reduce_dict(content_segmentation_masks, content_image))
cv2.imwrite(change_filename(args.seg_dir, args.style_image, '_seg', '.png'),
reduce_dict(style_segmentation_masks, style_image))
if args.init == "noise":
random_noise_scaling_factor = 0.0001
random_noise = np.random.randn(*content_image.shape).astype(np.float32)
if path is None or not os.path.isfile(path):
print("File {} does not exist.".format(path))
exit(0)
content_image = load_image(args.content_image)
# use existing if available
if (load_segmentation):
print("Load segmentation from files.")
content_segmentation_image = cv2.imread(
content_segmentation_filename)
content_segmentation_masks = extract_segmentation_masks(
content_segmentation_image)
else:
print("Create segmentation.")
content_segmentation = compute_segmentation(args.content_image)
cv2.imwrite(
change_filename(result_dir, args.content_image, '_seg_raw',
'.png'), content_segmentation)
content_segmentation_masks, color_to_gram_dict, anp_results = merge_anps(
content_segmentation, style_text, args.adjective_threshold,
args.noun_threshold, result_dir)
cv2.imwrite(
change_filename(result_dir, args.content_image, '_seg', '.png'),
reduce_dict(content_segmentation_masks, content_image))
if args.init == "noise":
random_noise_scaling_factor = 0.0001
random_noise = np.random.randn(*content_image.shape).astype(
from components.segmentation import compute_segmentation
from components.semantic_merge import extract_segmentation_masks
import itertools as it
from operator import itemgetter
import networkx as nx
import nltk
import numpy as np
import tensorflow as tf
from os.path import join
from components.PSPNet.model import load_color_label_dict
from components.path import WEIGHTS_DIR
content_segmentation, style_segmentation = compute_segmentation(
"/home/skhan22/PoeticStyleTransfer/data2/summer_lake/2108832-1366x768-[DesktopNexus.com].jpg",
"/home/skhan22/PoeticStyleTransfer/data2/summer_lake/2504-1271617411m0fL.jpg"
)
print(content_segmentation.shape)
print(style_segmentation.shape)
# load color - label mapping
color_label_dict = load_color_label_dict()
label_color_dict = {
label: color
for color, labels in color_label_dict.items() for label in labels
}
colors = color_label_dict.keys()
# Extract the boolean mask for every color
content_masks = extract_segmentation_masks(content_segmentation, colors)
def get_gram_from_generator(anp, color, save_dir, results, vgg19): #fake generator implementation
#just randomly picks an image from the ANP dataset and gets the gram matrices for it
image_names = []
for image_name in os.listdir(os.path.join(dataset_location, anp)):
if(image_name.endswith('.jpg')):
image_names.append(image_name)
rand_indices = list(range(len(image_names)))
random.shuffle(rand_indices)
selected_image = os.path.join(dataset_location.rstrip(), anp.rstrip(), image_names[rand_indices[0]].rstrip())
print('loading '+selected_image)
image_segmentation = compute_segmentation(selected_image)
image_masks = extract_segmentation_masks(image_segmentation)
i = 1
while not color in image_masks.keys():
selected_image = os.path.join(dataset_location, anp, image_names[rand_indices[i]].rstrip())
print('previous image didnt work loading' + selected_image)
image_segmentation = compute_segmentation(selected_image)
image_masks = extract_segmentation_masks(image_segmentation)
i += 1
# global first
# global first_selected_image
# if(first):
# first_selected_image = cv2.imread(selected_image)
# first = False
# colored_image = color_transfer(first_selected_image, cv2.imread(selected_image))
# cv2.imwrite(os.path.join(save_dir, anp+'.jpg'), colored_image)
# selected_image = os.path.join(save_dir, anp+'.jpg')
required_mask = mask_for_tf_with_color(image_masks, color)
results+=anp+' image: '+selected_image+'\n'
style_anp_image = load_image(selected_image)
#cv2.imwrite(os.path.join(save_dir, anp+'_seg_raw.png'), image_segmentation)
style_anp_image = vgg.preprocess(style_anp_image)
anp_gram_matrices = []
global weight_restorer
with tf.compat.v1.Session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
weight_restorer.init(sess)
style_convs = sess.run(
fetches=[vgg19.conv1_1, vgg19.conv2_1, vgg19.conv3_1, vgg19.conv4_1, vgg19.conv5_1],
feed_dict={image_placeholder: style_anp_image})
for style_conv in style_convs:
anp_gram_matrices.append(tf.compat.v1.identity(calculate_gram_matrix_with_mask(style_conv, required_mask)))
return anp_gram_matrices, results
