def __call__(self, img, gt):
img, gt = random_mirror(img, gt)
if config.train_scale_array is not None:
img, gt, scale = random_scale(img, gt, config.train_scale_array)
img = normalize(img, self.img_mean, self.img_std)
crop_size = (config.image_height, config.image_width)
crop_pos = generate_random_crop_pos(img.shape[:2], crop_size)
p_img, _ = random_crop_pad_to_shape(img, crop_pos, crop_size, 0)
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size, 0)
# scaled_gt = cv2.resize(p_gt, (
# config.image_width // 8, config.image_height // 8),
# interpolation=cv2.INTER_NEAREST)
#
# C = config.num_classes + 1
# one_hot_gt = convert_to_one_hot(scaled_gt.astype(np.int), C)
# similarity_gt = np.dot(one_hot_gt, one_hot_gt.T)
p_img = p_img.transpose(2, 0, 1)
# p_gt = p_gt - 1
extra_dict = None
return p_img, p_gt, extra_dict
def __call__(self, img, gt, edge, midline):
img, gt, edge, midline = random_mirror(img, gt, edge, midline)
gt = img_to_black(gt)
edge = img_to_black(edge)
midline = img_to_black(midline)
if config.train_scale_array is not None:
img, gt, scale, edge, midline = random_scale(
img, gt, config.train_scale_array, edge, midline)
img = normalize(img, self.img_mean, self.img_std)
p_img, p_gt, p_edge, p_midline = img, gt, edge, midline
p_img = cv2.resize(p_img,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_gt = cv2.resize(p_gt,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_edge = cv2.resize(p_edge,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_midline = cv2.resize(
p_midline, (config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
if p_gt.max() > 1:
print(p_gt)
return p_img, p_gt, extra_dict, p_edge, p_midline
def __call__(self, img, gt):
#flip
img, gt = random_mirror(img, gt)
#according the paper
if config.train_scale_array is not None:
img, gt, scale = random_scale(img, gt, config.train_scale_array)
id255 = np.where(gt == 255)
no255_gt = np.array(gt)
no255_gt[id255] = 0
cgt = cv2.Canny(no255_gt, 5, 5, apertureSize=7)
#get border imformation from canny
cgt = cv2.dilate(cgt, self.edge_kernel)
cgt[cgt == 255] = 1
#img white
img = normalize(img, self.img_mean, self.img_std)
crop_size = (config.image_height, config.image_width)
crop_pos = generate_random_crop_pos(img.shape[:2], crop_size)
p_img, _ = random_crop_pad_to_shape(img, crop_pos, crop_size, 0)
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size, 255)
p_cgt, _ = random_crop_pad_to_shape(cgt, crop_pos, crop_size, 255)
p_img = p_img.transpose(2, 0, 1)
extra_dict = {'aux_label': p_cgt}
return p_img, p_gt, extra_dict
def __call__(self, img, gt, edge, midline):
img, gt, edge, midline = random_mirror(
img, gt, edge,
midline) # images are randomly flipped to increase variance
gt = img_to_black(gt) # binary filter on gt.
edge = img_to_black(edge)
midline = img_to_black(midline)
if config.train_scale_array is not None:
img, gt, scale, edge, midline = random_scale(
img, gt, config.train_scale_array, edge,
midline) # scale the images with supplied list
img = normalize(img, self.img_mean, self.img_std)
crop_size = (200, 200)
crop_pos = generate_random_crop_pos(
img.shape[:2], crop_size) # obtain random location
p_img, _ = random_crop_pad_to_shape(
img, crop_pos, crop_size,
0) # get the cropped images and re-sized to crop-size
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size, -1)
p_edge, _ = random_crop_pad_to_shape(edge, crop_pos, crop_size, -1)
p_midline, _ = random_crop_pad_to_shape(midline, crop_pos, crop_size,
-1)
p_img = cv2.resize(
p_img, (config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST) # resize by downsampling
p_gt = cv2.resize(p_gt,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_edge = cv2.resize(p_edge,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_midline = cv2.resize(
p_midline, (config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
if p_gt.max() > 1:
print(p_gt)
return p_img, p_gt, extra_dict, p_edge, p_midline
def __call__(self, img, gt):
img, gt = random_mirror(img, gt)
if config.train_scale_array is not None:
img, gt, scale = random_scale(img, gt, config.train_scale_array)
img = normalize(img, self.img_mean, self.img_std)
crop_size = (config.image_height, config.image_width)
crop_pos = generate_random_crop_pos(img.shape[:2], crop_size)
p_img, _ = random_crop_pad_to_shape(img, crop_pos, crop_size, 0)
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size, 0)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
return p_img, p_gt, extra_dict
def __call__(self, img, gt):
img, gt = random_mirror(img, gt)
if self.config.train_scale_array is not None:
img, gt, scale = random_scale(img, gt, self.config.train_scale_array)
img = normalize(img, self.img_mean, self.img_std)
crop_size = (self.config.image_height, self.config.image_width)
crop_pos = generate_random_crop_pos(img.shape[:2], crop_size)
p_img, _ = random_crop_pad_to_shape(img, crop_pos, crop_size, 0)
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size, 255)
p_gt = cv2.resize(p_gt, (self.config.image_width // self.config.gt_down_sampling, self.config.image_height // self.config.gt_down_sampling), interpolation=cv2.INTER_NEAREST)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
return p_img, p_gt, extra_dict
def __call__(self, img, gt):
(img, gt) = random_mirror(img, gt)
gt = img_to_black(gt)
if config.train_scale_array is not None:
(img, gt, scale) = random_scale(img, gt, config.train_scale_array)
img = normalize(img, self.img_mean, self.img_std)
(p_img, p_gt) = (img, gt)
p_img = cv2.resize(p_img,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_gt = cv2.resize(p_gt,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
return (p_img, p_gt, extra_dict)
def __call__(self, img, gt):
img, gt = random_mirror(
img, gt) # images are randomly flipped to increase variance
gt = img_to_black(gt) # binary filter on gt.
if config.train_scale_array is not None:
img, gt, scale = random_scale(
img, gt, config.train_scale_array
) # scale the images with supplied list
img = normalize(img, self.img_mean, self.img_std)
crop_size = (200, 200)
crop_pos = generate_random_crop_pos(img.shape[:2], crop_size)
p_img, _ = random_crop_pad_to_shape(
img, crop_pos, crop_size,
0) # get the cropped images and re-sized to crop-size
p_gt, _ = random_crop_pad_to_shape(gt, crop_pos, crop_size,
255) # value=
p_img = cv2.resize(p_img,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_gt = cv2.resize(p_gt,
(config.image_width // config.gt_down_sampling,
config.image_height // config.gt_down_sampling),
interpolation=cv2.INTER_NEAREST)
p_img = p_img.transpose(2, 0, 1)
extra_dict = None
return p_img, p_gt, extra_dict
def __getitem__(self, index):
datafiles = self.files[index]
image = cv2.imread(datafiles["img"], cv2.IMREAD_COLOR)
label = cv2.imread(datafiles["label"], cv2.IMREAD_GRAYSCALE)
size = image.shape
name = datafiles["name"]
image = normalize(image, np.array(self.mean), np.array(self.std))
image, label = random_mirror(image, label)
if self.scale:
image, label, scale = random_scale(image, label,
[0.75, 1, 1.25, 1.5, 1.75, 2.0])
crop_pos = generate_random_crop_pos(image.shape[:2],
self.crop_size)
image, _ = random_crop_pad_to_shape(image, crop_pos,
self.crop_size, 0)
label, _ = random_crop_pad_to_shape(label, crop_pos,
self.crop_size, 255)
image = image.transpose(2, 0, 1)
return image.copy(), label.copy(), np.array(size), name
