def random_transform_group_entry(self, image, annotations):
# randomly transform both image and annotations
if self.transform_generator:
_, _, transform_mat = next(self.transform_generator)
transform = adjust_transform_for_image(transform_mat, image, self.transform_parameters.relative_translation)
image = apply_transform(transform, image, self.transform_parameters)
# Transform the bounding boxes in the annotations.
annotations = annotations.copy()
for index in range(annotations.shape[0]):
annotations[index, :4] = transform_aabb(transform, annotations[index, :4])
return image, annotations
def random_transform_group_entry(self, image, annotations):
"""
Randomly transforms image and annotation.
"""
# randomly transform both image and annotations
if transform is not None or self.transform_generator:
if transform is None:
transform = adjust_transform_for_image(next(self.transform_generator), image,
self.transform_parameters.relative_translation)
# apply transformation to image
image = apply_transform(transform, image, self.transform_parameters)
# Transform the bounding boxes in the annotations.
annotations['bboxes'] = annotations['bboxes'].copy()
for index in range(annotations['bboxes'].shape[0]):
annotations['bboxes'][index, :] = transform_aabb(transform, annotations['bboxes'][index, :])
return image, annotations
def random_transform_group_entry(self, image, annotations, transform=None):
"""!@brief
Randomly transforms image and annotation.
"""
#-------------------------------------------
# # Draw grid lines for print/debug purposes
# grid_size = 20
# for i in range(0, image.shape[1], grid_size):
# cv2.line(image, (i, 0), (i, image.shape[0]), color=(50,50,50))
# for j in range(0, image.shape[0], grid_size):
# cv2.line(image, (0, j), (image.shape[1], j), color=(50,50,50))
#
# # Draw the annotations
# from utils.visualization import draw_annotations
# draw_annotations(image,
# annotations,
# color=(0, 0, 255))
#
# # Show input image
# cv2.imshow("Input", image)
#-------------------------------------------
# Randomly apply photometric distortions
if self.photometric:
# Apply photometric distortion to the image
image, annotations = photometric_distortions(image, annotations)
# Randomly apply point-spread-function (PSF)
# blurring to simulate animal motion
if self.motion:
# Create a random PSF kernel
kernel = createPSFKernel(image.shape)
# Kernel size can be small enough (< 3)
# to not apply the motion blur.
if kernel is not None:
# Apply PSF blur to the image
image = apply_motion_blur(image, kernel)
# Adjust bounding boxes accordingly
annotations['bboxes'] = annotations['bboxes'].copy()
for index in range(annotations['bboxes'].shape[0]):
annotations['bboxes'][index, :] = motion_distortion_aabb(
image.shape, kernel, annotations['bboxes'][index, :])
#-------------------------------------------
# # Draw the annotations
# draw_annotations(image,
# annotations,
# color=(255, 0, 0))
# cv2.imshow("PSF_Blur", image)
#-------------------------------------------
# Randomly apply smooth elastic deformation to the image
if self.deformable:
# Get deformable array
deformable_transform = adjust_deformable_for_image(
image.shape, self.alpha, self.sigma)
# Apply transformation to the image
image = apply_deformable_transform(image, deformable_transform)
# Transform the bounding boxes in the annotations
annotations['bboxes'] = annotations['bboxes'].copy()
n_bboxes = annotations['bboxes'].shape[0]
# Non parallelized option
for index in range(n_bboxes):
annotations['bboxes'][index, :] = deformable_transform_aabb(
image.shape, deformable_transform,
annotations['bboxes'][index, :])
## Parallelized option
## See: https://www.machinelearningplus.com/python/parallel-processing-python/
## Init multiprocessing.Pool()
#pool = mp.Pool(mp.cpu_count())
#results = [pool.apply_async(deformable_transform_aabb,
#args=(image.shape,
#deformable_transform,
#annotations['bboxes'][i, :])
#) for i in range(n_bboxes)]
## Close Pool and let all the processes complete
#pool.close()
#pool.join() # postpones the execution of next line of code until all processes in the queue are done.
## <results> is a list of pool.ApplyResult objects
## NOTE: This attribution is not ORDERED since async
## is being used to parallelize the function
#for i in range(n_bboxes):
#annotations['bboxes'][i, :] = results[i].get()
#-------------------------------------------
# # Draw the annotations
# draw_annotations(image,
# annotations,
# color=(255, 0, 0))
# cv2.imshow("Middle", image)
#-------------------------------------------
# Randomly transform both image and annotations
if transform is not None or self.transform_generator:
# Affine + flip transformation
if transform is None:
transform = adjust_transform_for_image(
next(self.transform_generator), image,
self.transform_parameters.relative_translation)
# Apply transformation to the image
image = apply_transform(transform, image,
self.transform_parameters)
# Transform the bounding boxes in the annotations
annotations['bboxes'] = annotations['bboxes'].copy()
for index in range(annotations['bboxes'].shape[0]):
annotations['bboxes'][index, :] = transform_aabb(
transform, annotations['bboxes'][index, :])
#-------------------------------------------
# # Draw the annotations
# from utils.visualization import draw_annotations
# draw_annotations(image,
# annotations,
# color=(0, 255, 0))
# # Show output image after data augmentation
# cv2.imshow("Output", image)
# cv2.waitKey(0)
#-------------------------------------------
return image, annotations
