def train_transforms(sample, image_shape, jittering):
"""
Training data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
jittering : tuple (brightness, contrast, saturation, hue)
Color jittering parameters
Returns
-------
sample : dict
Augmented sample
"""
if len(image_shape) > 0:
sample = resize_sample(sample, image_shape)
sample = duplicate_sample(sample)
if len(jittering) > 0:
sample = colorjitter_sample(sample, jittering)
sample = to_tensor_sample(sample)
return sample
def train_transforms(sample, image_shape, jittering, **kwargs):
"""
Training data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
jittering : tuple (brightness, contrast, saturation, hue)
Color jittering parameters
Returns
-------
sample : dict
Augmented sample
"""
# (orig_w, orig_h) = sample['rgb'].size
# (h, w) = shape
# factor = w/orig_w
# h=h*factor
if len(image_shape) > 0:
sample = resize_sample(sample, image_shape)
sample = duplicate_sample(sample)
if len(jittering) > 0:
sample = colorjitter_sample(sample, jittering)
if "max_roll_angle" in kwargs:
sample = rotate_sample(sample, degrees=kwargs["max_roll_angle"])
if "random_center_crop" in kwargs:
if kwargs["random_center_crop"]:
sample = random_center_crop_sample(sample)
sample = to_tensor_sample(sample)
return sample
def test_transforms(sample, image_shape, crop_eval_borders):
"""
Test data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
Returns
-------
sample : dict
Augmented sample
"""
if len(crop_eval_borders) > 0:
borders = parse_crop_borders(crop_eval_borders,
sample['rgb'].size[::-1])
sample = crop_sample_input(sample, borders)
if len(image_shape) > 0:
sample['rgb'] = resize_image(sample['rgb'], image_shape)
if 'input_depth' in sample:
sample['input_depth'] = resize_depth(sample['input_depth'],
image_shape)
sample = to_tensor_sample(sample)
return sample
def validation_transforms(sample, image_shape, crop_eval_borders):
"""
Validation data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
crop_eval_borders : tuple (left, top, right, down)
Border for cropping
Returns
-------
sample : dict
Augmented sample
"""
if len(crop_eval_borders) > 0:
borders = parse_crop_borders(crop_eval_borders,
sample['rgb'].size[::-1])
sample = crop_sample_input(sample, borders)
if len(image_shape) > 0:
sample['rgb'] = resize_image(sample['rgb'], image_shape)
if 'input_depth' in sample:
sample['input_depth'] = resize_depth_preserve(
sample['input_depth'], image_shape)
sample = to_tensor_sample(sample)
return sample
def train_transforms(sample, image_shape, jittering, crop_train_borders):
"""
Training data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
jittering : tuple (brightness, contrast, saturation, hue)
Color jittering parameters
crop_train_borders : tuple (left, top, right, down)
Border for cropping
Returns
-------
sample : dict
Augmented sample
"""
if len(crop_train_borders) > 0:
borders = parse_crop_borders(crop_train_borders,
sample['rgb'].size[::-1])
sample = crop_sample(sample, borders)
if len(image_shape) > 0:
sample = resize_sample(sample, image_shape)
sample = duplicate_sample(sample)
if len(jittering) > 0:
sample = colorjitter_sample(sample, jittering)
sample = to_tensor_sample(sample)
return sample
def test_transforms(sample, image_shape):
"""
Test data augmentation transformations
Parameters
----------
sample : dict
Sample to be augmented
image_shape : tuple (height, width)
Image dimension to reshape
Returns
-------
sample : dict
Augmented sample
"""
if len(image_shape) > 0:
sample['rgb'] = resize_image(sample['rgb'], image_shape)
sample = to_tensor_sample(sample)
return sample
