import random
import cv2
import numpy as np
def crop(img, offeset):
h1, w1, h2, w2 = offeset
return img[h1:h2, w1:w2, ...]
def random_crop(img, target_size):
h, w = img.shape[0:2]
th, tw = target_size
h1 = random.randint(0, max(0, h - th))
w1 = random.randint(0, max(0, w - tw))
h2 = min(h1 + th, h)
w2 = min(w1 + tw, w)
return crop(img, [h1, w1, h2, w2])
def group_random_crop(img_group, target_size):
h, w = img_group[0].shape[0:2]
th, tw = target_size
h1 = random.randint(0, max(0, h - th))
w1 = random.randint(0, max(0, w - tw))
h2 = min(h1 + th, h)
w2 = min(w1 + tw, w)
outs = list()
for img in img_group:
assert (img.shape[0] == h and img.shape[1] == w)
outs.append(crop(img, [h1, w1, h2, w2]))
return outs
def center_crop(img, target_size):
h, w = img.shape[0:2]
th, tw = target_size
h1 = max(0, int((h - th) / 2))
w1 = max(0, int((w - tw) / 2))
h2 = min(h1 + th, h)
w2 = min(w1 + tw, w)
return crop(img, [h1, w1, h2, w2])
def group_center_crop(img_group, target_size):
h, w = img_group[0].shape[0:2]
th, tw = target_size
h1 = max(0, int((h - th) / 2))
w1 = max(0, int((w - tw) / 2))
h2 = min(h1 + th, h)
w2 = min(w1 + tw, w)
outs = list()
for img in img_group:
assert (img.shape[0] == h and img.shape[1] == w)
outs.append(crop(img, [h1, w1, h2, w2]))
return outs
def pad(img, offeset, value=0):
h1, w1, h2, w2 = offeset
img = cv2.copyMakeBorder(
img, h1, h2, w1, w2, cv2.BORDER_CONSTANT, value=value)
return img
def random_pad(img, target_size, value=0):
h, w = img.shape[0:2]
th, tw = target_size
h1 = random.randint(0, max(0, th - h))
w1 = random.randint(0, max(0, tw - w))
h2 = max(th - h - h1, 0)
w2 = max(tw - w - w1, 0)
return pad(img, [h1, w1, h2, w2], value=value)
def group_random_pad(img_group, target_size, values):
h, w = img_group[0].shape[0:2]
th, tw = target_size
h1 = random.randint(0, max(0, th - h))
w1 = random.randint(0, max(0, tw - w))
h2 = max(th - h - h1, 0)
w2 = max(tw - w - w1, 0)
outs = list()
for img, value in zip(img_group, values):
assert (img.shape[0] == h and img.shape[1] == w)
outs.append(pad(img, [h1, w1, h2, w2], value=value))
return outs
def center_pad(img, target_size, value=0):
h, w = img.shape[0:2]
th, tw = target_size
h1 = max(0, int((th - h) / 2))
w1 = max(0, int((tw - w) / 2))
h2 = max(th - h - h1, 0)
w2 = max(tw - w - w1, 0)
return pad(img, [h1, w1, h2, w2], value=value)
def group_center_pad(img_group, target_size, values):
h, w = img_group[0].shape[0:2]
th, tw = target_size
h1 = max(0, int((th - h) / 2))
w1 = max(0, int((tw - w) / 2))
h2 = max(th - h - h1, 0)
w2 = max(tw - w - w1, 0)
outs = list()
for img, value in zip(img_group, values):
assert (img.shape[0] == h and img.shape[1] == w)
outs.append(pad(img, [h1, w1, h2, w2], value=value))
return outs
def group_concer_pad(img_group, target_size, values):
h, w = img_group[0].shape[0:2]
th, tw = target_size
h1 = 0
w1 = 0
h2 = max(th - h - h1, 0)
w2 = max(tw - w - w1, 0)
outs = list()
for img, value in zip(img_group, values):
assert (img.shape[0] == h and img.shape[1] == w)
outs.append(pad(img, [h1, w1, h2, w2], value=value))
return outs
def rescale(img, scales, interpolation=cv2.INTER_LINEAR):
if isinstance(scales, list):
if len(scales) == 2:
scale = random.uniform(scales[0], scales[1])
else:
scale = random.choice(scales)
else:
scale = scales
img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=interpolation)
return img
def group_rescale(img_group, scales, interpolations):
if isinstance(scales, list):
if len(scales) == 2:
scale = random.uniform(scales[0], scales[1])
else:
scale = random.choice(scales)
else:
scale = scales
outs = list()
for img, interpolation in zip(img_group, interpolations):
outs.append(rescale(img, scale, interpolation))
return outs
def rotation(img, degrees, interpolation=cv2.INTER_LINEAR, value=0):
if isinstance(degrees, list):
if len(degrees) == 2:
degree = random.uniform(degrees[0], degrees[1])
else:
degree = random.choice(degrees)
else:
degree = degrees
h, w = img.shape[0:2]
center = (w / 2, h / 2)
map_matrix = cv2.getRotationMatrix2D(center, degree, 1.0)
img = cv2.warpAffine(
img,
map_matrix, (w, h),
flags=interpolation,
borderMode=cv2.BORDER_CONSTANT,
borderValue=value)
return img
def group_rotation(img_group, degrees, interpolations, values):
if isinstance(degrees, list):
if len(degrees) == 2:
degree = random.uniform(degrees[0], degrees[1])
else:
degree = random.choice(degrees)
else:
degree = degrees
outs = list()
for img, interpolation, value in zip(img_group, interpolations, values):
outs.append(rotation(img, degree, interpolation, value))
return outs
def flip(img):
return np.fliplr(img)
def random_flip(img):
if random.random() < 0.5:
return flip(img)
else:
return img
def group_flip(img_group):
return [flip(img) for img in img_group]
def group_random_flip(img_group):
if random.random() < 0.5:
return [flip(img) for img in img_group]
else:
return img_group
def normalize(img, mean, std=None):
img = img - np.array(mean)[np.newaxis, np.newaxis, ...]
if std is not None:
img = img / np.array(std)[np.newaxis, np.newaxis, ...]
return img
def blur(img, kenrel_size=(5, 5), sigma=(1e-6, 0.6)):
img = cv2.GaussianBlur(img, kenrel_size, random.uniform(*sigma))
return img
def random_blur(img, kenrel_size=(5, 5), sigma=(1e-6, 0.6)):
if random.random() < 0.5:
return blur(img, kenrel_size, sigma)
else:
return img
