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 apply(self, samples):
# Apply the transformation and return the transformed version.
# sample must be a 3D tensor with shape [rows,cols,channels].
out = np.zeros(samples.shape)
for i in range(len(samples)):
out[i] = samples[i]
if self.shiftX > 0 or self.shiftY > 0:
out[i] = image.random_shift(out[i], self.shiftX, self.shiftY)
if self.rot > 0:
out[i] = image.random_rotation(out[i], self.rot)
if self.shear > 0:
out[i] = image.random_shear(out[i], self.shear)
if self.zoomMin > 0 and self.zoomMax > 0:
out[i] = image.random_zoom(out[i],
[self.zoomMin, self.zoomMax])
if self.flipH and np.random.uniform() < 0.5:
out[i] = image.flip_axis(out[i], 1)
if self.flipV and np.random.uniform() < 0.5:
out[i] = image.flip_axis(out[i], 2)
return out
def shear_image(img_arr, givenIntensity=5):
return kim.random_shear(img_arr,
intensity=givenIntensity,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
def test_random_transforms(self):
x = np.random.random((2, 28, 28))
assert image.random_rotation(x, 45).shape == (2, 28, 28)
assert image.random_shift(x, 1, 1).shape == (2, 28, 28)
assert image.random_shear(x, 20).shape == (2, 28, 28)
assert image.random_zoom(x, (5, 5)).shape == (2, 28, 28)
assert image.random_channel_shift(x, 20).shape == (2, 28, 28)
def rnd_shear(img, factor):
return random_shear(img,
intensity=factor,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
def augmentImages(batch_of_images, batch_of_masks):
for i in range(len(batch_of_images)):
img_and_mask = np.concatenate((batch_of_images[i, ...], batch_of_masks[i, ...]), axis=2)
if img_and_mask.ndim == 4: # This assumes single channel data. For multi-channel you'll need
# change this to put all channel in slices channel
orig_shape = img_and_mask.shape
img_and_mask = img_and_mask.reshape((img_and_mask.shape[0:3]))
if np.random.randint(0, 10) == 7:
img_and_mask = random_rotation(img_and_mask, rg=45, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:
img_and_mask = random_shift(img_and_mask, wrg=0.2, hrg=0.2, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:
# 没任何显示效果，f**k
img_and_mask = random_shear(img_and_mask, intensity=15, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='nearest', cval=0.)
if np.random.randint(0, 10) == 7:
# fill_mode = 'nearest'
# zoom_range=(0.1, 0.5) #超近距离
# zoom_range=(0.5, 5) # 超远距离
zoom_range = (0.35, 2) # 一般般
img_and_mask = random_zoom(img_and_mask, zoom_range=zoom_range, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:#按y轴左右翻转
img_and_mask = flip_axis(img_and_mask, axis=1)
# if np.random.randint(0, 10) == 7:#按x轴上下翻转，上下倒过来了有点恐怖
# img_and_mask = flip_axis(img_and_mask, axis=0)
#椒盐噪声
# if np.random.randint(0, 10) == 7:
# # amount=10 #密密麻麻
# amount = 2 #正常些
# salt_pepper_noise(img_and_mask, salt=1, amount=amount)
# 这个elastic_transform导致显示的东西像屎一样，干脆不要了
# if np.random.randint(0, 5) == 3:
# img_and_mask = elastic_transform(img_and_mask, alpha=1000, sigma=80, alpha_affine=50)
#最终的维度处理
if batch_of_images.ndim == 4:
batch_of_images[i, ...] = img_and_mask[..., 0:3]
batch_of_masks[i, ...] = np.expand_dims(img_and_mask[...,3],axis=-1)
# Ensure the masks did not get any non-binary values.
batch_of_masks[batch_of_masks > 0.5] = 1
batch_of_masks[batch_of_masks <= 0.5] = 0
return (batch_of_images, batch_of_masks)
def augmentation_pipeline(img_arr):
import keras.preprocessing.image as im
img_arr = im.random_rotation(img_arr,
18,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = im.random_shear(img_arr,
intensity=0.4,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = im.random_zoom(img_arr,
zoom_range=(0.9, 2.0),
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = random_greyscale(img_arr, 0.4)
return img_arr
def test_random_transforms(self):
x = np.random.random((2, 28, 28))
assert image.random_rotation(x, 45).shape == (2, 28, 28)
assert image.random_shift(x, 1, 1).shape == (2, 28, 28)
assert image.random_shear(x, 20).shape == (2, 28, 28)
assert image.random_zoom(x, (5, 5)).shape == (2, 28, 28)
assert image.random_channel_shift(x, 20).shape == (2, 28, 28)
def augment_data(dataset, dataset_labels, augmentation_factor=1, use_random_rotation=True, use_random_shear=True, use_random_shift=True, use_random_zoom=True):
augmented_image = []
augmented_image_labels = []
for num in range (0, dataset.shape[0]):
# original image:
augmented_image.append(dataset[num])
augmented_image_labels.append(dataset_labels[num])
for i in range(0, augmentation_factor):
if use_random_rotation:
augmented_image.append(random_rotation(dataset[num], 20, row_axis=0, col_axis=1, channel_axis=2))
augmented_image_labels.append(dataset_labels[num])
if use_random_shear:
augmented_image.append(random_shear(dataset[num], 0.2, row_axis=0, col_axis=1, channel_axis=2))
augmented_image_labels.append(dataset_labels[num])
if use_random_shift:
augmented_image.append(random_shift(dataset[num], 0.2, 0.2, row_axis=0, col_axis=1, channel_axis=2))
augmented_image_labels.append(dataset_labels[num])
s = np.arange(len(augmented_image))#Permutation
return np.array(augmented_image)[s], np.array(augmented_image_labels)[s]
def img_augmentation(img, num):
img_arr_origin = img_to_array(img)
img_list = []
for i in range(num):
img_arr = random_rotation(img_arr_origin,
40,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = random_shear(img_arr,
intensity=0.4,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = random_zoom(img_arr,
zoom_range=(0.9, 1.2),
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img = array_to_img(img_arr)
img = np.array(img)
img_list.append(img)
#img_arr = random_greyscale(img_arr, 0.4)
return img_list
def paint_text(text, w, h, rotate=False, move=False, multi_fonts=False, background=False):
surface = cairo.ImageSurface(cairo.FORMAT_RGB24, w, h)
with cairo.Context(surface) as context:
context.set_source_rgb(1, 1, 1) # White
context.paint()
if multi_fonts:
# Calibri Century Comic Sans Courier New Futura Georgia
fonts = ['Century Schoolbook', 'Courier', 'Arial', 'STIX','Tahoma','Times New Roman','Trebuchet MS',
'Verdana','Wide Latin','Calibri','Century','Comic Sans','Courier','New Futura','Georgia',
'Lucida','Lucida Console','Magneto','Mistral','URW Chancery L', 'FreeMono','DejaVue Sans Mono']
font_slant = np.random.choice([cairo.FONT_SLANT_NORMAL,cairo.FONT_SLANT_ITALIC,cairo.FONT_SLANT_OBLIQUE])
font_weight = np.random.choice([cairo.FONT_WEIGHT_BOLD, cairo.FONT_WEIGHT_NORMAL, cairo.FONT_WEIGHT_NORMAL])
context.select_font_face(np.random.choice(fonts), font_slant, font_weight)
else:
context.select_font_face('Courier', cairo.FONT_SLANT_NORMAL, cairo.FONT_WEIGHT_BOLD)
# context.set_font_size(25)
font_size = randint(12, 42)
context.set_font_size(font_size)
box = context.text_extents(text)
border_w_h = (font_size/2, font_size/2)
# if box[2] > (w - 2 * border_w_h[1]) or box[3] > (h - 2 * border_w_h[0]):
# raise IOError('Could not fit string into image. Max char count is too large for given image width.')
# teach the RNN translational invariance by
# fitting text box randomly on canvas, with some room to rotate
min_x = 0 #font_size/4
min_y = 0# font_size/4
max_shift_x = w - box[2] - border_w_h[0]
max_shift_y = h - box[3] - border_w_h[1]
if max_shift_x <= min_x :
top_left_x = 10
else:
top_left_x = np.random.randint(min_x, int(max_shift_x))
if move and max_shift_y > min_y + 1:
top_left_y = np.random.randint(min_y, int(max_shift_y))
else:
top_left_y = h // 2
context.move_to(top_left_x - int(box[0]), top_left_y - int(box[1]))
context.set_source_rgb(0, 0, 0)
if text:
context.show_text(text)
buf = surface.get_data()
a = np.frombuffer(buf, np.uint8)
a.shape = (h, w, 4)
a = a[:, :, 0] # grab single channel
a = a.astype(np.float32) / 255
a = np.expand_dims(a, 0)
if rotate:
angle = randint(0, 3)
a = image.random_rotation(a, angle * (w - top_left_x) / w + 1)
sheer = randint(0, 3)
a = image.random_shear(a,sheer)
if background:
a = speckle(a)
return a
def augment_image(self, img_arr):
img_arr = random_rotation(img_arr, 18, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest')
img_arr = random_shear(img_arr, intensity=0.4, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest')
img_arr = random_zoom(img_arr, zoom_range=(0.9, 2.0), row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest')
if img_arr.shape[-1] == 3:
img_arr = self.random_greyscale(img_arr, 0.4)
return img_arr
def show_preprocessed_example(X_train):
x = X_train[5]
row_axis, col_axis, channel_axis = 0, 1, 2
# random rotation
x_r = io.random_rotation(x,
rg=45,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
# random horizontal shift
x_ws = io.random_shift(x,
wrg=.2,
hrg=0,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
# random vertical shift
x_hs = io.random_shift(x,
wrg=0,
hrg=.2,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
# random shear
x_s = io.random_shear(x,
intensity=.2,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
# random zoom
x_z = io.random_zoom(x,
zoom_range=(.8, 1.2),
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
images = [x, x_r, x_ws, x_hs, x_s, x_z]
titles = [
"Original", "Rotate", "Horizontal shift", "Vertical shift", "Shear",
"Zoom"
]
plt.figure(figsize=(8, 4.5))
for i, image in enumerate(images):
plt.subplot(2, 3, i + 1)
plt.xticks(())
plt.yticks(())
plt.title(titles[i])
image = inputs.global_contrast_normalization(image)
image = inputs.minmax_normalization(image)
plt.imshow(image)
plt.show()
def transform(x, seed=0):
np.random.seed(seed)
img = image_processing.random_rotation(x, 0.2)
img = image_processing.random_shear(img, 30)
img = image_processing.random_zoom(img, (0.5, 1.1))
if np.random.rand() >= 0.5:
img = np.fliplr(img)
return img
def augment_data(self, x):
x = random_shift(x, self.width_shift_range, self.height_shift_range,
row_axis=0, col_axis = 1, channel_axis = 2)
x = random_rotation(x, self.max_rotation,
row_axis = 0, col_axis = 1, channel_axis = 2)
x = random_shear(x, self.shear, row_axis = 0, col_axis = 1, channel_axis = 2)
x = random_zoom(x, self.zoom_range, row_axis = 0, col_axis = 1, channel_axis = 2)
return x
def test_img_transforms(self):
x = np.random.random((3, 200, 200))
_ = preprocessing_image.random_rotation(x, 20)
_ = preprocessing_image.random_shift(x, 0.2, 0.2)
_ = preprocessing_image.random_shear(x, 2.)
_ = preprocessing_image.random_zoom(x, (0.5, 0.5))
_ = preprocessing_image.apply_channel_shift(x, 2, 2)
_ = preprocessing_image.apply_affine_transform(x, 2)
with self.assertRaises(ValueError):
preprocessing_image.random_zoom(x, (0, 0, 0))
_ = preprocessing_image.random_channel_shift(x, 2.)
def data_augmentor(x,mode,row_axis=1,col_axis=0,channel_axis=-1):
#temp = [0,0,0,0,0,0]
#for i in range(len(mode)):
#temp[6-i-1] = int(str(mode[-i-1]))
if int(mode[0]):
x = flip_axis(x, 0)
#print x.shape
if int(mode[1]):
x = flip_axis(x,1)
#print x.shape
if int(mode[2]):
x = random_rotation(x,360, row_axis, col_axis, channel_axis,fill_mode='reflect')
x = np.swapaxes(x,0,1)
x = np.swapaxes(x,1,2)
'''M = cv2.getRotationMatrix2D((x.shape[1],x.shape[0]),np.random.randint(360),1) #last argument is scale in rotation
x = cv2.warpAffine(x,M,(x.shape[1],x.shape[0]), borderMode=cv2.BORDER_REFLECT)'''
#print x.shape
#del M
if int(mode[3]):
x = random_shift(x, 0.1,0.1, row_axis, col_axis, channel_axis,fill_mode='reflect')
x = np.swapaxes(x,0,1)
x = np.swapaxes(x,1,2)
'''M = np.float32([[1,0,np.random.randint(x.shape[0])],[0,1,np.random.randint(x.shape[1])]])
x = cv2.warpAffine(x,M,(x.shape[1],x.shape[0]), borderMode = cv2.BORDER_REFLECT)'''
#print x.shape
#del M
if int(mode[4]):
x = random_shear(x, 1, row_axis, col_axis, channel_axis,fill_mode='reflect')
x = np.swapaxes(x,0,1)
x = np.swapaxes(x,1,2)
'''pts1 = np.float32([[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])],[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])],[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])]])
pts2 = np.float32([[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])],[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])],[np.random.randint(x.shape[0]),np.random.randint(x.shape[1])]])
M = cv2.getAffineTransform(pts1,pts2)
x = cv2.warpAffine(x,M,(x.shape[1],x.shape[0]),borderMode = cv2.BORDER_REFLECT)'''
#print x.shape
#del M
#del pts1
#del pts2
if int(mode[5]):
x = random_zoom(x, (1.1,1.1), row_axis, col_axis, channel_axis,fill_mode='reflect')
x = np.swapaxes(x,0,1)
x = np.swapaxes(x,1,2)
#print x.shape
return x
def test_random_transforms(self):
x = np.random.random((2, 28, 28))
assert image.random_rotation(x, 45).shape == (2, 28, 28)
assert image.random_shift(x, 1, 1).shape == (2, 28, 28)
assert image.random_shear(x, 20).shape == (2, 28, 28)
assert image.random_zoom(x, (5, 5)).shape == (2, 28, 28)
assert image.random_channel_shift(x, 20).shape == (2, 28, 28)
# Test get_random_transform with predefined seed
seed = 1
generator = image.ImageDataGenerator(
rotation_range=90.,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.5,
zoom_range=0.2,
channel_shift_range=0.1,
brightness_range=(1, 5),
horizontal_flip=True,
vertical_flip=True)
transform_dict = generator.get_random_transform(x.shape, seed)
transform_dict2 = generator.get_random_transform(x.shape, seed * 2)
assert transform_dict['theta'] != 0
assert transform_dict['theta'] != transform_dict2['theta']
assert transform_dict['tx'] != 0
assert transform_dict['tx'] != transform_dict2['tx']
assert transform_dict['ty'] != 0
assert transform_dict['ty'] != transform_dict2['ty']
assert transform_dict['shear'] != 0
assert transform_dict['shear'] != transform_dict2['shear']
assert transform_dict['zx'] != 0
assert transform_dict['zx'] != transform_dict2['zx']
assert transform_dict['zy'] != 0
assert transform_dict['zy'] != transform_dict2['zy']
assert transform_dict['channel_shift_intensity'] != 0
assert (transform_dict['channel_shift_intensity'] !=
transform_dict2['channel_shift_intensity'])
assert transform_dict['brightness'] != 0
assert transform_dict['brightness'] != transform_dict2['brightness']
# Test get_random_transform without any randomness
generator = image.ImageDataGenerator()
transform_dict = generator.get_random_transform(x.shape, seed)
assert transform_dict['theta'] == 0
assert transform_dict['tx'] == 0
assert transform_dict['ty'] == 0
assert transform_dict['shear'] == 0
assert transform_dict['zx'] == 1
assert transform_dict['zy'] == 1
assert transform_dict['channel_shift_intensity'] is None
assert transform_dict['brightness'] is None
def test_random_transforms(self):
x = np.random.random((2, 28, 28))
assert image.random_rotation(x, 45).shape == (2, 28, 28)
assert image.random_shift(x, 1, 1).shape == (2, 28, 28)
assert image.random_shear(x, 20).shape == (2, 28, 28)
assert image.random_zoom(x, (5, 5)).shape == (2, 28, 28)
assert image.random_channel_shift(x, 20).shape == (2, 28, 28)
# Test get_random_transform with predefined seed
seed = 1
generator = image.ImageDataGenerator(
rotation_range=90.,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.5,
zoom_range=0.2,
channel_shift_range=0.1,
brightness_range=(1, 5),
horizontal_flip=True,
vertical_flip=True)
transform_dict = generator.get_random_transform(x.shape, seed)
transform_dict2 = generator.get_random_transform(x.shape, seed * 2)
assert transform_dict['theta'] != 0
assert transform_dict['theta'] != transform_dict2['theta']
assert transform_dict['tx'] != 0
assert transform_dict['tx'] != transform_dict2['tx']
assert transform_dict['ty'] != 0
assert transform_dict['ty'] != transform_dict2['ty']
assert transform_dict['shear'] != 0
assert transform_dict['shear'] != transform_dict2['shear']
assert transform_dict['zx'] != 0
assert transform_dict['zx'] != transform_dict2['zx']
assert transform_dict['zy'] != 0
assert transform_dict['zy'] != transform_dict2['zy']
assert transform_dict['channel_shift_intensity'] != 0
assert (transform_dict['channel_shift_intensity'] !=
transform_dict2['channel_shift_intensity'])
assert transform_dict['brightness'] != 0
assert transform_dict['brightness'] != transform_dict2['brightness']
# Test get_random_transform without any randomness
generator = image.ImageDataGenerator()
transform_dict = generator.get_random_transform(x.shape, seed)
assert transform_dict['theta'] == 0
assert transform_dict['tx'] == 0
assert transform_dict['ty'] == 0
assert transform_dict['shear'] == 0
assert transform_dict['zx'] == 1
assert transform_dict['zy'] == 1
assert transform_dict['channel_shift_intensity'] is None
assert transform_dict['brightness'] is None
def generator(samples, batch_size=32):
"""
This function defination generates training features and lables for given samples and batch size
samples - list of lists that stores images, steering angles, speed etc.,
parse through the list and read images (center, left and right camera) and steering angles with (0, 0.2, -0.2) corrections
images are augmented using following methods
- image flip
- addition of noise
- random shear
"""
num_samples = len(samples)
while 1: # Loop forever so the generator never terminates
for offset in range(0, num_samples, batch_size):
batch_samples = samples[offset:offset + batch_size]
images = []
angles = []
for batch_sample in batch_samples:
for i, correction in zip(range(3), corrections):
# read image and steering angle + correction (depending upon center, left or right camera image)
name = './data/IMG/' + batch_sample[i].split('\\')[-1]
image = mpimg.imread(name)
angle = float(batch_sample[3]) + correction
# original image
images.append(image)
angles.append(angle)
# augmentation
# flip image
images.append(np.fliplr(image))
angles.append(angle * -1.0)
# add noise
images.append(image - 0.35)
angles.append(angle)
# random shear
images.append(random_shear(image, np.random.randint(10)))
angles.append(angle)
# trim image to only see section with road
X_train = np.array(images)
y_train = np.array(angles)
yield shuffle(X_train, y_train)
def random_transform(x,
seed=42,
rotation_range=30,
width_shift_range=.1,
height_shift_range=.1,
shear_range=0.2,
zoom_range=(.9, 1.1),
row_axis=0,
col_axis=1,
channel_axis=2):
'''
https://github.com/keras-team/keras/blob/master/keras/preprocessing/image.py
:param x: 3D input image
:param seed:
:param rotation_range: Rotation range, in degrees.
:param width_shift_range: Width shift range, as a float fraction of the width.
:param height_shift_range: Height shift range, as a float fraction of the height.
:param shear_range: Transformation intensity
:param zoom_range: Tuple of floats; zoom range for width and height.
:return: transformed image that has the same shape
'''
random.seed(seed)
x = io.random_rotation(x,
rotation_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_shift(x,
width_shift_range,
height_shift_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_shear(x,
shear_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_zoom(x,
zoom_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
return x
def original_process(original_img):
# 随机旋转
process_image = image.random_rotation(original_img,
180,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='constant',
cval=0)
# 随机投影变换
h, w = IMAGE_SIZE[0], IMAGE_SIZE[1]
ratio = np.random.normal(0.125, 0.075)
if ratio > 0.2:
ratio = 0.2
elif ratio < 0.05:
ratio = 0.05
pts1 = np.float32([[0, 0], [w, 0], [0, h], [w, h]])
pts2 = np.float32([[0, 0], [w, ratio * h], [0, h],
[w, (1 - ratio) * h]])
M = cv2.getPerspectiveTransform(pts1, pts2)
process_image = cv2.warpPerspective(process_image, M, (w, h))
# 随机平移
process_image = image.random_shift(process_image,
0.3,
0.3,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='constant',
cval=0)
# 随机拉伸
ratio = np.random.normal(35, 10)
if ratio > 45:
ratio = 45
elif ratio < 25:
ratio = 25
process_image = image.random_shear(process_image,
ratio,
row_axis=0,
col_axis=1,
channel_axis=2)
return process_image
def data_augmentation(imgs, c):
# imgs - list of 2D images
def sequence_change(imgs, change_faktor):
# change_faktor: float from 0 to 0.4
l = random.randint(0, int(len(imgs) * change_faktor))
r = random.randint(0, int(len(imgs) * change_faktor)) + 1
return imgs[l:-r]
imgs = sequence_change(imgs, c.sequence_change)
imgs = np.stack(imgs)
imgs = resample_imgs(imgs, c.n_frames)
imgs = image.random_zoom(imgs, c.zoom_range)
imgs = image.random_rotation(imgs, c.random_rotation)
imgs = image.random_shear(imgs, c.random_shear)
return imgs
def augmentImages(batch_of_images, batch_of_masks):
for i in range(len(batch_of_images)):
img_and_mask = np.concatenate((batch_of_images[i, ...], batch_of_masks[i, ...]), axis=2)
if img_and_mask.ndim == 4: # This assumes single channel data. For multi-channel you'll need
# change this to put all channel in slices channel
orig_shape = img_and_mask.shape
img_and_mask = img_and_mask.reshape((img_and_mask.shape[0:3]))
if np.random.randint(0, 10) == 7:
img_and_mask = random_rotation(img_and_mask, rg=45, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
# if np.random.randint(0, 5) == 3:
# img_and_mask = elastic_transform(img_and_mask, alpha=1000, sigma=80, alpha_affine=50)
#这个elastic_transform导致显示的东西像屎一样，干脆不要了
if np.random.randint(0, 10) == 7:
img_and_mask = random_shift(img_and_mask, wrg=0.2, hrg=0.2, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:
img_and_mask = random_shear(img_and_mask, intensity=16, row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:
img_and_mask = random_zoom(img_and_mask, zoom_range=(0.75, 0.75), row_axis=0, col_axis=1, channel_axis=2,
fill_mode='constant', cval=0.)
if np.random.randint(0, 10) == 7:
img_and_mask = flip_axis(img_and_mask, axis=1)
if np.random.randint(0, 10) == 7:
img_and_mask = flip_axis(img_and_mask, axis=0)
if np.random.randint(0, 10) == 7:
salt_pepper_noise(img_and_mask, salt=0.2, amount=0.04)
if batch_of_images.ndim == 4:
batch_of_images[i, ...] = img_and_mask[..., 0:3]
batch_of_masks[i, ...] = np.expand_dims(img_and_mask[...,3],axis=-1)
# Ensure the masks did not get any non-binary values.
batch_of_masks[batch_of_masks > 0.5] = 1
batch_of_masks[batch_of_masks <= 0.5] = 0
return (batch_of_images, batch_of_masks)
def __getitem__(self, idx):
batch_x = self.x[idx * self.batch_size:(idx + 1) * self.batch_size]
batch_y = self.y[idx * self.batch_size:(idx + 1) * self.batch_size]
X = []
Y = []
for file_name, y in zip(batch_x, batch_y):
img = create_image(file_name, self.shape)
if img is None:
continue
img = self.horizontal_flip(img)
img = random_rotation(img, 0.20)
img = random_shift(img, 0.10, 0.10)
img = random_shear(img, 0.10)
X.append(img)
Y.append(y)
X = np.array(X).astype('float32') / 255.0
return np.array(X), np.array(Y)
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 augmentation_pipeline(img_arr):
img_arr = random_rotation(img_arr,
18,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = random_shear(img_arr,
intensity=0.4,
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
img_arr = random_zoom(img_arr,
zoom_range=(0.9, 2.0),
row_axis=0,
col_axis=1,
channel_axis=2,
fill_mode='nearest')
return img_arr
def argumentation(imgs):
print(np.shape(imgs))
r_imgs = [
random_rotation(imgs,
30,
row_axis=1,
col_axis=2,
channel_axis=3,
fill_mode='nearest') * 255 for _ in range(5)
]
s_imgs = [
random_shear(imgs,
intensity=0.4,
row_axis=1,
col_axis=2,
channel_axis=3,
fill_mode='nearest') * 255 for _ in range(5)
]
sh_imgs = [
random_shift(imgs,
wrg=0.1,
hrg=0.3,
row_axis=1,
col_axis=2,
channel_axis=3,
fill_mode='nearest') * 255 for _ in range(5)
]
z_imgs = [
random_zoom(imgs,
zoom_range=(1.5, 0.7),
row_axis=1,
col_axis=2,
channel_axis=3,
fill_mode='nearest') * 255 for _ in range(5)
]
imgs = np.append(imgs, r_imgs)
imgs = np.append(imgs, s_imgs)
imgs = np.append(imgs, sh_imgs)
imgs = np.append(imgs, z_imgs)
return imgs
def random_transform(x, seed=42, rotation_range=30,
width_shift_range=.1, height_shift_range=.1,
shear_range=0.2, zoom_range = (.9, 1.1),
row_axis=0, col_axis=1, channel_axis=2):
'''
https://github.com/keras-team/keras/blob/master/keras/preprocessing/image.py
:param x: 3D input image
:param seed:
:param rotation_range: Rotation range, in degrees.
:param width_shift_range: Width shift range, as a float fraction of the width.
:param height_shift_range: Height shift range, as a float fraction of the height.
:param shear_range: Transformation intensity
:param zoom_range: Tuple of floats; zoom range for width and height.
:return: transformed image that has the same shape
'''
random.seed(seed)
x = io.random_rotation(x, rotation_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_shift(x, width_shift_range,
height_shift_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_shear(x, shear_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
x = io.random_zoom(x, zoom_range,
row_axis=row_axis,
col_axis=col_axis,
channel_axis=channel_axis)
return x
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 generator(samples, batch_size=32):
num_samples = len(samples)
while 1:
shuffle(samples)
for offset in range(0, num_samples, batch_size):
batch_samples = samples[offset:offset + batch_size]
images = []
angles = []
for batch_sample in batch_samples:
for i, correction in zip(range(3), angle_correction):
name = './data/udacity/IMG/' + batch_sample[i].split(
'/')[-1]
image = mpimg.imread(name)
angle = float(batch_sample[3]) + correction
images.append(image)
angles.append(angle)
#data augumentation:
#flip image
images.append(np.fliplr(image))
angles.append(-angle)
# data agumentation: random shear
images.append(random_shear(image, np.random.randint(32)))
angles.append(angle)
# convert to np array
X_train = np.array(images)
y_train = np.array(angles)
yield shuffle(X_train, y_train)
def image_augmentation(img):
'''
Arguments:
img = list of np.arrays of images
Returns:
list of arrays after transformations
'''
rotationSize = 15
zoomRangeWidth = .8
zoomRangeHeight = 1.1
intensity = 15
widthRange = .1
heightRange = .1
img = [kim.random_rotation(item, rotationSize, row_axis=0, col_axis=1,
channel_axis=2, fill_mode='nearest') for item in img]
img = [kim.random_shift(item, wrg=widthRange, hrg=heightRange,row_axis=0,
col_axis=1, channel_axis=2, fill_mode='nearest') for item in img]
img = [kim.random_zoom(item, zoom_range=(zoomRangeWidth,zoomRangeHeight),
row_axis=0,col_axis=1, channel_axis=2, fill_mode='nearest') for item in img]
img = [kim.random_shear(item, intensity=intensity,
row_axis=0,col_axis=1, channel_axis=2, fill_mode='nearest') for item in img]
return img
def generator(samples, batch_size=32):
num_samples = len(samples)
while 1:
for offset in range(0, num_samples, batch_size):
batch_samples = samples[offset:offset + batch_size]
images = []
angles = []
for batch_sample in batch_samples:
for i, correction in zip(range(3), angle_offsets):
name = './data/all/IMG/' + batch_sample[i].split('/')[-1]
#print (name)
image = mpimg.imread(name)
angle = float(batch_sample[3]) + correction
# append original image
images.append(image)
angles.append(angle)
# data agumentation: flip image (invert angle)
images.append(np.fliplr(image))
angles.append(angle * -1.0)
# data agumentation: random shear
images.append(random_shear(image, np.random.randint(32)))
angles.append(angle)
# convert to np array
X_train = np.array(images)
y_train = np.array(angles)
# Shuffle the data
yield shuffle(X_train, y_train)
