def augment(self,
n,
hflip=False,
vflip=False,
scale_to_percent=1.0,
scale_axis_equally=True,
rotation_deg=0,
shear_deg=0,
translation_x_px=0,
translation_y_px=0,
brightness_change=0.0,
noise_mean=0.0,
noise_std=0.0):
"""Generates randomly augmented versions of the image.
Also augments the keypoints accordingly.
Args:
n Number of augmentations to generate.
hflip Allow horizontal flipping (yes/no).
vflip Allow vertical flipping (yes/no)
scale_to_percent How much scaling/zooming to allow. Values are around 1.0.
E.g. 1.1 is -10% to +10%
E.g. (0.7, 1.05) is -30% to 5%.
scale_axis_equally Whether to enforce equal scaling of x and y axis.
rotation_deg How much rotation to allow. E.g. 5 is -5 degrees to +5 degrees.
shear_deg How much shearing to allow.
translation_x_px How many pixels of translation along the x axis to allow.
translation_y_px How many pixels of translation along the y axis to allow.
brightness_change How much change in brightness to allow. Values are around 0.0.
E.g. 0.2 is -20% to +20%.
noise_mean Mean value of gaussian noise to add.
noise_std Standard deviation of gaussian noise to add.
Returns:
List of ImageWithKeypoints
"""
assert n >= 0
result = []
if n == 0:
return result
matrices = create_aug_matrices(n,
img_width_px=self.get_width(),
img_height_px=self.get_height(),
scale_to_percent=scale_to_percent,
scale_axis_equally=scale_axis_equally,
rotation_deg=rotation_deg,
shear_deg=shear_deg,
translation_x_px=translation_x_px,
translation_y_px=translation_y_px)
for i in range(n):
img = self.copy()
matrix = matrices[i]
# random horizontal / vertical flip
if hflip and random.random() > 0.5:
img.image_arr = np.fliplr(img.image_arr)
img.keypoints.fliplr(img)
if vflip and random.random() > 0.5:
img.image_arr = np.flipud(img.image_arr)
img.keypoints.flipud(img)
# random brightness adjustment
by_percent = random.uniform(1.0 - brightness_change,
1.0 + brightness_change)
img.image_arr = img.image_arr * by_percent
# gaussian noise
# numpy requires a std above 0
if noise_std > 0:
img.image_arr = img.image_arr \
+ (255 * np.random.normal(noise_mean, noise_std,
(img.image_arr.shape)))
# clip to 0-255
img.image_arr = np.clip(img.image_arr, 0, 255).astype(np.uint8)
arr = tf.warp(img.image_arr, matrix,
mode="constant") # projects to float 0-1
img.image_arr = np.array(arr * 255, dtype=np.uint8)
img.keypoints.warp(img, matrix)
result.append(img)
return result
def augment(
self,
n,
hflip=False,
vflip=False,
scale_to_percent=1.0,
scale_axis_equally=True,
rotation_deg=0,
shear_deg=0,
translation_x_px=0,
translation_y_px=0,
brightness_change=0.0,
noise_mean=0.0,
noise_std=0.0,
):
"""Generates randomly augmented versions of the image.
Also augments the keypoints accordingly.
Args:
n Number of augmentations to generate.
hflip Allow horizontal flipping (yes/no).
vflip Allow vertical flipping (yes/no)
scale_to_percent How much scaling/zooming to allow. Values are around 1.0.
E.g. 1.1 is -10% to +10%
E.g. (0.7, 1.05) is -30% to 5%.
scale_axis_equally Whether to enforce equal scaling of x and y axis.
rotation_deg How much rotation to allow. E.g. 5 is -5 degrees to +5 degrees.
shear_deg How much shearing to allow.
translation_x_px How many pixels of translation along the x axis to allow.
translation_y_px How many pixels of translation along the y axis to allow.
brightness_change How much change in brightness to allow. Values are around 0.0.
E.g. 0.2 is -20% to +20%.
noise_mean Mean value of gaussian noise to add.
noise_std Standard deviation of gaussian noise to add.
Returns:
List of ImageWithKeypoints
"""
assert n >= 0
result = []
if n == 0:
return result
matrices = create_aug_matrices(
n,
img_width_px=self.get_width(),
img_height_px=self.get_height(),
scale_to_percent=scale_to_percent,
scale_axis_equally=scale_axis_equally,
rotation_deg=rotation_deg,
shear_deg=shear_deg,
translation_x_px=translation_x_px,
translation_y_px=translation_y_px,
)
for i in range(n):
img = self.copy()
matrix = matrices[i]
# random horizontal / vertical flip
if hflip and random.random() > 0.5:
img.image_arr = np.fliplr(img.image_arr)
img.keypoints.fliplr(img)
if vflip and random.random() > 0.5:
img.image_arr = np.flipud(img.image_arr)
img.keypoints.flipud(img)
# random brightness adjustment
by_percent = random.uniform(1.0 - brightness_change, 1.0 + brightness_change)
img.image_arr = img.image_arr * by_percent
# gaussian noise
# numpy requires a std above 0
if noise_std > 0:
img.image_arr = img.image_arr + (255 * np.random.normal(noise_mean, noise_std, (img.image_arr.shape)))
# clip to 0-255
img.image_arr = np.clip(img.image_arr, 0, 255).astype(np.uint8)
arr = tf.warp(img.image_arr, matrix, mode="nearest") # projects to float 0-1
img.image_arr = np.array(arr * 255, dtype=np.uint8)
img.keypoints.warp(img, matrix)
result.append(img)
return result
def main():
"""Measure time required to generate augmentations matrices and to apply
them.
"""
batch_size = 64
nb_runs = 20
# Measure time required to generate 100k augmentation matrices
"""
print("Generating 100 times 1000 augmentation matrices of size 64x64...")
start = time.time()
for _ in range(100):
create_aug_matrices(1000, 64, 64,
scale_to_percent=1.5, scale_axis_equally=False,
rotation_deg=20, shear_deg=20,
translation_x_px=5, translation_y_px=5)
print("Done in %.8f" % (time.time() - start,))
"""
# Test Performance on 64 images of size 512x512 pixels
image = data.lena()
images = np.resize(
image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[0],
image.shape[1],
hflip=True,
vflip=True,
scale_to_percent=1.3,
scale_axis_equally=False,
rotation_deg=25,
shear_deg=10,
translation_x_px=5,
translation_y_px=5)
print("Running tests on %d images of shape %s" %
(batch_size, str(image.shape)))
run_tests(augmenter, images, nb_runs)
print("")
print("Running tests on %d images of shape %s" %
(batch_size, str(image.shape)))
print("(With 1000 pregenerated matrices)")
augmenter.pregenerate_matrices(1000)
run_tests(augmenter, images, nb_runs)
print("")
# Test Performance on 64 images of size 64x64 pixels
image = data.lena()
image = misc.imresize(image, (64, 64))
images = np.resize(
image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[0],
image.shape[1],
hflip=True,
vflip=True,
scale_to_percent=1.3,
scale_axis_equally=False,
rotation_deg=25,
shear_deg=10,
translation_x_px=5,
translation_y_px=5)
print("Running tests on %d images of shape %s" %
(batch_size, str(image.shape)))
run_tests(augmenter, images, nb_runs)
print("Running tests on %d images of shape %s" %
(batch_size, str(image.shape)))
print("(With 1000 pregenerated matrices)")
augmenter.pregenerate_matrices(1000)
run_tests(augmenter, images, nb_runs)
print("")
# Time required to augment 1,000,000 images of size 32x32
print("Augmenting 1000 batches of 1000 lena images (1 million total)" \
", each of size 32x32...")
image = data.lena()
image = misc.imresize(image, (32, 32))
batch_size = 1000
images = np.resize(
image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[1],
image.shape[0],
hflip=True,
vflip=True,
scale_to_percent=1.3,
scale_axis_equally=False,
rotation_deg=25,
shear_deg=10,
translation_x_px=5,
translation_y_px=5)
augmenter.pregenerate_matrices(1000)
start = time.time()
for _ in range(1000):
augmenter.augment_batch(images)
print("Done in %.8fs" % (time.time() - start, ))
print("")
# Time required to augment 1,000,000 images of size 32x32
# but using only one matrix without the class (no library overhead from
# ImageAugmenter)
# Notice that this does not include horizontal and vertical flipping,
# which is done via numpy in the ImageAugmenter class.
print("Augmenting 1000 batches of 1000 lena images (1 million total)" \
", each of size 32x32, using one matrix directly (no ImageAugmenter " \
"class)...")
matrices = create_aug_matrices(1,
image.shape[1],
image.shape[0],
scale_to_percent=1.3,
scale_axis_equally=False,
rotation_deg=25,
shear_deg=10,
translation_x_px=5,
translation_y_px=5)
matrix = matrices[0]
start = time.time()
for _ in range(1000):
for image in images:
augmented_image = tf.warp(image, matrix)
print("Done in %.8fs" % (time.time() - start, ))
def main():
"""Measure time required to generate augmentations matrices and to apply
them.
"""
batch_size = 64
nb_runs = 20
# Measure time required to generate 100k augmentation matrices
"""
print("Generating 100 times 1000 augmentation matrices of size 64x64...")
start = time.time()
for _ in range(100):
create_aug_matrices(1000, 64, 64,
scale_to_percent=1.5, scale_axis_equally=False,
rotation_deg=20, shear_deg=20,
translation_x_px=5, translation_y_px=5)
print("Done in %.8f" % (time.time() - start,))
"""
# Test Performance on 64 images of size 512x512 pixels
image = data.lena()
images = np.resize(image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[0], image.shape[1],
hflip=True, vflip=True,
scale_to_percent=1.3, scale_axis_equally=False,
rotation_deg=25, shear_deg=10,
translation_x_px=5, translation_y_px=5)
print("Running tests on %d images of shape %s" % (batch_size, str(image.shape)))
run_tests(augmenter, images, nb_runs)
print("")
print("Running tests on %d images of shape %s" % (batch_size, str(image.shape)))
print("(With 1000 pregenerated matrices)")
augmenter.pregenerate_matrices(1000)
run_tests(augmenter, images, nb_runs)
print("")
# Test Performance on 64 images of size 64x64 pixels
image = data.lena()
image = misc.imresize(image, (64, 64))
images = np.resize(image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[0], image.shape[1],
hflip=True, vflip=True,
scale_to_percent=1.3, scale_axis_equally=False,
rotation_deg=25, shear_deg=10,
translation_x_px=5, translation_y_px=5)
print("Running tests on %d images of shape %s" % (batch_size, str(image.shape)))
run_tests(augmenter, images, nb_runs)
print("Running tests on %d images of shape %s" % (batch_size, str(image.shape)))
print("(With 1000 pregenerated matrices)")
augmenter.pregenerate_matrices(1000)
run_tests(augmenter, images, nb_runs)
print("")
# Time required to augment 1,000,000 images of size 32x32
print("Augmenting 1000 batches of 1000 lena images (1 million total)" \
", each of size 32x32...")
image = data.lena()
image = misc.imresize(image, (32, 32))
batch_size = 1000
images = np.resize(image, (batch_size, image.shape[0], image.shape[1], image.shape[2]))
augmenter = ImageAugmenter(image.shape[1], image.shape[0],
hflip=True, vflip=True,
scale_to_percent=1.3, scale_axis_equally=False,
rotation_deg=25, shear_deg=10,
translation_x_px=5, translation_y_px=5)
augmenter.pregenerate_matrices(1000)
start = time.time()
for _ in range(1000):
augmenter.augment_batch(images)
print("Done in %.8fs" % (time.time() - start,))
print("")
# Time required to augment 1,000,000 images of size 32x32
# but using only one matrix without the class (no library overhead from
# ImageAugmenter)
# Notice that this does not include horizontal and vertical flipping,
# which is done via numpy in the ImageAugmenter class.
print("Augmenting 1000 batches of 1000 lena images (1 million total)" \
", each of size 32x32, using one matrix directly (no ImageAugmenter " \
"class)...")
matrices = create_aug_matrices(1, image.shape[1], image.shape[0],
scale_to_percent=1.3, scale_axis_equally=False,
rotation_deg=25, shear_deg=10,
translation_x_px=5, translation_y_px=5)
matrix = matrices[0]
start = time.time()
for _ in range(1000):
for image in images:
augmented_image = tf.warp(image, matrix)
print("Done in %.8fs" % (time.time() - start,))
