def augmentation_img(img1,img2):
aug = np.random.randint(2,size=5)
img_a1, img_a2 = np.zeros_like(img1) , np.zeros_like(img2)
if aug[0]:
# 旋转
img_a1 = image.random_rotation(img1,30,0,1,2)
img_a2 = image.random_rotation(img2,30,0,1,2)
if aug[1]:
# 错位
img_a1 = image.random_shear(img1,30,0,1,2)
img_a2 = image.random_shear(img2,30,0,1,2)
if aug[2]:
# 缩放
img_a1 = image.random_zoom(img_a1,(0.8,1.2),0,1,2)
img_a2 = image.random_zoom(img_a2,(0.8,1.2),0,1,2)
if aug[3]:
# 亮度
img_a1 = image.random_brightness(img_a1, (0.5,1.5))
if aug[4]:
# 噪声
img_a1 += np.random.uniform(0,32,(np.shape(img_a1)))
img_a1 = np.clip(img_a1,0,255)
# 转成 0-255 整数
img_a1 = np.array(img_a1,'uint8')
img_a2 = np.array(img_a2,'uint8')
return img_a1, img_a2
def augmentation_img(img1,img2):
'''
图像增强（旋转/翻转/缩放/亮度调整/噪声）
'''
aug = np.random.randint(2,size=5)
img_a1, img_a2 = np.copy(img1) , np.copy(img2)
if aug[0]:# 旋转
angle = np.random.choice((1,2,-1,-2))
img_a1 = np.rot90(img_a1,angle)
img_a2 = np.rot90(img_a2,angle)
if aug[1]:# 对称翻转
x_or_y = np.random.choice((0,1))
img_a1 = np.flip(img_a1,x_or_y)
img_a2 = np.flip(img_a2,x_or_y)
if aug[2]:# 缩放
zoom = np.random.uniform(0.8,1.2)
img_a1 = image.random_zoom(img_a1,(zoom,zoom),0,1,2)
img_a2 = image.random_zoom(img_a2,(zoom,zoom),0,1,2)
if aug[3]:# 亮度
img_a1 = image.random_brightness(img_a1, (0.5,1.5))
if aug[4]:# 噪声
img_a1 += np.random.uniform(0,8,(np.shape(img_a1)))
img_a1 = np.clip(img_a1,0,255)
# 转成 0-255 整数
img_a1 = np.array(img_a1,'uint8')
img_a2 = np.array(img_a2,'uint8')
return img_a1, img_a2
def tran_one(img, d=15):
degree = np.random.randint(-d, d)
img = iprocess.random_brightness(img, (0.5, 1.5))
if np.random.rand() >= 0.5:
img = np.fliplr(img)
return iprocess.apply_affine_transform(img, degree)
def __getitem__(self, index):
data = self.data[index * self.batch_size: (index + 1) * self.batch_size]
words = []
images = []
nums = []
for i in range(self.batch_size):
img_path = data[i][0][0]
word = data[i][1][0]
word = self.__word_process(word)
img = get_img("{}/{}".format(self.path, img_path), self.width, self.height)
img = random_brightness(img[:, :, np.newaxis], [0.1, 1.5])
images.append(img[:, :, 0])
words.append(word)
nums.append(len(word))
words = pad_sequences(words, maxlen=self.max_string_length, truncating="post", padding="post")
label_input = np.array(nums).reshape([-1, 1])
images = np.array(images)
return {"img": images[:, :, :, np.newaxis], "label_input": label_input,
"y_input": words}, np.zeros([self.batch_size, 1])
def augm(array):
flag = int(random.randint(0, 6) / 7)
if flag == 0:
a = array
if flag == 1:
a = image.random_shift(array, -0.2, 0.2)
if flag == 2:
a = image.random_shear(array, 80)
if flag == 3:
a = image.random_zoom(array, (0.7, 1))
if flag == 4:
a = img_pca(array)
if flag == 5:
a = image.random_brightness(array, (0.05, 0.8))
if flag == 6:
a = image.random_rotation(array, 180)
# if flag == 7:
# a = image.random_channel_shift(array, 0.05)
return a
def image_augment(arr):
"""
:param arr: array of a image, 3D. [array]
:return:
"""
# 旋转 微调
x = image.random_rotation(arr,
rg=5,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
# 平移 上下平移 左右平移 微调
x = image.random_shift(x,
wrg=0.1,
hrg=0.1,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
# 随机空间剪切
x = image.random_shear(x,
intensity=10,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
# 随机放缩
x = image.random_zoom(x,
zoom_range=(0.75, 1.25),
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
# 随机亮度调整
x = image.random_brightness(x, brightness_range=(0.75, 1.25))
return x
def random_transform(self, x, seed=None):
"""Randomly augments a single image tensor.
Rotation and image flips are applied to both satellite and roadmap.
Channel shifts and brightness changes are applied to satellite only.
# Arguments
x: 3D tensor, single image.
seed: Random seed.
# Returns
A randomly transformed version of the input (same shape).
"""
# x is a single image, so it doesn't have image number at index 0
img_row_axis = self.row_axis - 1
img_col_axis = self.col_axis - 1
img_channel_axis = self.channel_axis - 1
if seed is not None:
np.random.seed(seed)
# Use composition of homographies
# to generate final transform that needs to be applied
if self.rotation_range:
theta = np.deg2rad(
np.random.uniform(-self.rotation_range, self.rotation_range))
else:
theta = 0
transform_matrix = None
if theta != 0:
rotation_matrix = np.array([[np.cos(theta), -np.sin(theta), 0],
[np.sin(theta),
np.cos(theta), 0], [0, 0, 1]])
transform_matrix = rotation_matrix
if transform_matrix is not None:
h, w = x.shape[img_row_axis], x.shape[img_col_axis]
transform_matrix = transform_matrix_offset_center(
transform_matrix, h, w)
x = apply_transform(x,
transform_matrix,
img_channel_axis,
fill_mode=self.fill_mode,
cval=self.cval)
if self.horizontal_flip:
if np.random.random() < 0.5:
x = flip_axis(x, img_col_axis)
if self.vertical_flip:
if np.random.random() < 0.5:
x = flip_axis(x, img_row_axis)
x = np.moveaxis(x, img_channel_axis, -1)
x_satellite = x[:, :, 0:3]
x_roadmap = x[:, :, 3:6]
if self.channel_shift_range != 0:
x_satellite = random_channel_shift(x_satellite,
self.channel_shift_range,
img_channel_axis)
if self.brightness_range is not None:
x_satellite = random_brightness(x_satellite, self.brightness_range)
x = np.concatenate([x_satellite, x_roadmap], axis=-1)
x = np.moveaxis(x, -1, img_channel_axis)
return x