def test():
from color_transfer.img import DAY, NIGHT
src_file = DAY
ref_file = NIGHT
src = imread(src_file)
ref = imread(ref_file)
patch_match_c = True
interp = 2 # bilinear
normalize = "standardize"
level = 1
t = ADE20KTransfer(patch_match_c)
transferred = t.transfer(src, ref, level, interp, normalize)
display([src, ref, transferred])
def test():
from color_transfer.img import DAY, NIGHT
# choose your source and reference images
src_file, ref_file = DAY, NIGHT
src = imread(src_file)
ref = imread(ref_file)
# parameters, normally no need to change
patch_match_c = True
interp = 2 # bilinear
normalize = "standardize"
level = 1
n_clusters = 3 # 3 for sky, building and ground
total_iter = 5 # number of iterations for PatchMatch
epochs = 250 # number of epochs for transfer estimate
# color transfer
ct = ClusterTransfer(patch_match_c)
transferred = ct.transfer(src, ref, level, n_clusters, interp, normalize, total_iter, epochs)
display([src, ref, transferred])
def test():
import os
import matplotlib.pyplot as plt
from color_transfer.parts.pilutil import imread
path = "/".join(
os.path.realpath(__file__).replace(
"\\", "/").split("/")[:-1]) + "/parts/test/pspnet50_ade20k"
pattern = path + "/%d.jpg"
weights = "E:/ai/weights/pretrained/pspnet50_ade20k.h5"
sr = SkyRecognizer(weights)
for i in range(10):
file = pattern % i
img = imread(file)
mask = sr.recognize(img)
ax = plt.subplot(121)
ax.imshow(img)
ax = plt.subplot(122)
ax.imshow(mask)
plt.show()
import os
import numpy as np
import matplotlib.pyplot as plt
from color_transfer.parts.pilutil import imread, imresize
from color_transfer.parts.fast_guided_filter import fast_guided_filter
path = "/".join(os.path.realpath(__file__).replace("\\", "/").split("/")[:-1])
a = imread(path + "/a.jpg")
b = imread(path + "/b.jpg")
w, h = a.shape[:2]
w, h = w // 8, h // 8
a_low = imresize(a, (w, h)) / 255
b_low = imresize(b, (w, h)) / 255
a_high = a / 255
# equal channels
b_high = fast_guided_filter(a_low, b_low, a_high)
plt.subplot(221).imshow(a_low)
plt.subplot(222).imshow(a_high)
plt.subplot(223).imshow(b_low)
plt.subplot(224).imshow(b_high)
plt.show()
# different channels
a_low = np.mean(a_low, axis=2, keepdims=True)
a_high = np.mean(a_high, axis=2, keepdims=True)
b_high = fast_guided_filter(a_low, b_low, a_high)
plt.subplot(221).imshow(a_low[:, :, 0])
plt.subplot(222).imshow(a_high[:, :, 0])
plt.subplot(223).imshow(b_low)
plt.subplot(224).imshow(b_high)
plt.show()
import os
import numpy as np
import matplotlib.pyplot as plt
from color_transfer.parts.pilutil import imread
from color_transfer.parts.nnf_computation_c import nn_search, bds_vote
path = "/".join(os.path.realpath(__file__).replace("\\", "/").split("/")[:-1])
src = imread(path + "/a.jpg")
ref = imread(path + "/b.jpg")
patch_size = 3
total_iter = 2
w = 1
nnf_sr, nnf_rs = nn_search(src, ref, patch_size, total_iter) # nn search
guide = bds_vote(ref, nnf_sr, nnf_rs, patch_size, w) # bds vote
guide = np.uint8(guide)
plt.subplot(121).imshow(src)
plt.subplot(122).imshow(guide)
plt.show()
import os
import matplotlib.pyplot as plt
from color_transfer.parts.pspnet50_ade20k import PSPNet50ADE20K
from color_transfer.parts.pilutil import imread
path = "/".join(os.path.realpath(__file__).replace("\\", "/").split("/")[:-1])
pattern = path + "/%d.jpg"
psp = PSPNet50ADE20K()
for i in range(10):
file = pattern % i
img = imread(file)
predicted = psp.colorize(psp.predict(img))
ax = plt.subplot(121)
ax.imshow(img)
ax = plt.subplot(122)
ax.imshow(predicted)
plt.show()