def _aug_image(self, instance, net_h, net_w):
image_name = instance['filename']
image_name = image_name.replace('.raw', '.mhd')
image = raw_reader(image_name, instance['img_num'])
if image.shape[0] != 512:
return None, None
if image is None:
print('Cannot find ', image_name)
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w
dh = self.jitter * image_h
# dw = 0
# dh = 0
new_ar = (image_w + np.random.normal(0, dw)) / (
image_h + np.random.normal(0, dh))
# scale = np.random.uniform(0.25, 2)
scale = np.random.normal(1, 0.15)
if new_ar < 1:
new_h = int(scale * net_h)
new_w = int(net_h * new_ar)
else:
new_w = int(scale * net_w)
new_h = int(net_w / new_ar)
dx = int(np.random.uniform(0, net_w - new_w))
dy = int(np.random.uniform(0, net_h - new_h))
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w,
net_h, dx, dy)
# randomly distort hsv space
# im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
# correct the size and pos of bounding boxes
all_objs = correct_bounding_boxes(instance['object'], new_w, new_h,
net_w, net_h, dx, dy, flip, image_w,
image_h)
im_sized = self.gauss_noise(im_sized, 0, 0.001)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
# Read image in BGR format
filename = instance['filename']
image = cv2.imread(filename)
if image is None:
raise RuntimeError("Unable to load image file: %s" % filename)
# Convert to RGB
image = image[:,:,::-1]
# Apply jitter and scaling
image_h, image_w, _ = image.shape
dw = self.aug_jitter * image_w
dh = self.aug_jitter * image_h
new_ar = (image_w + np.random.uniform(-dw, dw)) / (image_h + np.random.uniform(-dh, dh))
scale = np.random.uniform(self.aug_scale[0], self.aug_scale[1])
if (new_ar < 1):
new_h = int(scale * net_h)
new_w = int(new_h * new_ar)
else:
new_w = int(scale * net_w)
new_h = int(new_w / new_ar)
# Apply padding to place the image in a random position within the net
if not self.aug_pad:
dx = 0
dy = 0
else:
dx = int(np.random.uniform(0, net_w - new_w))
dy = int(np.random.uniform(0, net_h - new_h))
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w, net_h, dx, dy)
# randomly distort hsv space
im_sized = random_distort_image(im_sized, self.aug_hue, self.aug_saturation, self.aug_exposure)
# randomly flip
if self.aug_flip:
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
else:
flip = False
# Make into gray image
if self.aug_gray:
im_sized = cv2.cvtColor(im_sized, cv2.COLOR_RGB2GRAY)[:,:,np.newaxis]
# correct the size and pos of bounding boxes
all_objs = correct_bounding_boxes(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, image_w, image_h)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
image_name = instance['filename']
image_name = image_name.replace(
'yolo_v3', '') # hack for changed folder structure
image = cv2.imread(image_name) # RGB image
if image is None: print('Cannot find ', image_name)
image = image[:, :, ::-1] # RGB image
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w
dh = self.jitter * image_h
new_ar = (image_w + np.random.uniform(-dw, dw)) / (
image_h + np.random.uniform(-dh, dh))
scale = np.random.uniform(0.95, 1.05)
if (new_ar < 1):
new_h = int(scale * net_h)
new_w = int(net_h * new_ar)
else:
new_w = int(scale * net_w)
new_h = int(net_w / new_ar)
dx = int(np.random.uniform(0, net_w - new_w))
dy = int(np.random.uniform(0, net_h - new_h))
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w,
net_h, dx, dy)
# randomly distort hsv space
# im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
#im_sized = random_flip(image, flip)
flip2 = np.random.randint(2)
im_sized = random_flip2(im_sized, flip2)
# correct the size and pos of bounding boxes
#all_objs = correct_bounding_boxes(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, image_w, image_h)
all_objs = correct_bounding_boxes2(instance['object'], new_w, new_h,
net_w, net_h, dx, dy, flip, flip2,
image_w, image_h)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
#image_name = "." + instance['filename']
image_name = instance['filename']
pil_image = PIL.Image.open(image_name).convert('RGB')
image = np.array(pil_image)
#image = cv2.imread(image_name) # RGB image
if image is None: print('Cannot find ', image_name)
#'image = image[:, :, ::-1] # RGB image
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w
dh = self.jitter * image_h
new_ar = (image_w + np.random.uniform(-dw, dw)) / (
image_h + np.random.uniform(-dh, dh))
scale = np.random.uniform(0.25, 2)
if (new_ar < 1):
new_h = int(scale * net_h)
new_w = int(net_h * new_ar)
else:
new_w = int(scale * net_w)
new_h = int(net_w / new_ar)
dx = int(np.random.uniform(0, net_w - new_w))
dy = int(np.random.uniform(0, net_h - new_h))
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w,
net_h, dx, dy)
# randomly distort hsv space
im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
# correct the size and pos of bounding boxes
all_objs = correct_bounding_boxes(instance['object'], new_w, new_h,
net_w, net_h, dx, dy, flip, image_w,
image_h)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
_, image_name = os.path.split(instance['filename'])
full_image_name = self.im_dir + image_name
image = cv2.imread(full_image_name) # RGB image
if image is None: print('Cannot find ', full_image_name)
image = image[:,:,::-1] # RGB image
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w;
dh = self.jitter * image_h;
new_ar = (image_w + np.random.uniform(-dw, dw)) / (image_h + np.random.uniform(-dh, dh));
scale = np.random.uniform(0.95, 1.05);
if (new_ar < 1):
new_h = int(scale * net_h);
new_w = int(net_h * new_ar);
else:
new_w = int(scale * net_w);
new_h = int(net_w / new_ar);
dx = int(np.random.uniform(0, net_w - new_w));
dy = int(np.random.uniform(0, net_h - new_h));
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w, net_h, dx, dy)
# randomly distort hsv space
# im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
#im_sized = random_flip(image, flip)
#flip2 = np.random.randint(2)
#im_sized = random_flip2(im_sized, flip2)
# correct the size and pos of bounding boxes
all_objs = correct_bounding_boxes(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, image_w, image_h)
#all_objs = correct_bounding_boxes2(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, flip2, image_w, image_h)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
_, image_name = os.path.split(instance['filename'])
full_image_name = self.im_dir + image_name
image = cv2.imread(full_image_name) # RGB image
if image is None: print('Cannot find ', full_image_name)
image = image[:,:,::-1] # RGB image
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w;
dh = self.jitter * image_h;
new_ar = (image_w + np.random.uniform(-dw, dw)) / (image_h + np.random.uniform(-dh, dh));
scale = np.random.uniform(0.95, 1.05);
if (new_ar < 1):
new_h = int(scale * net_h);
new_w = int(net_h * new_ar);
else:
new_w = int(scale * net_w);
new_h = int(net_w / new_ar);
dx = int(np.random.uniform(0, net_w - new_w));
dy = int(np.random.uniform(0, net_h - new_h));
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w, net_h, dx, dy)
# randomly distort hsv space
# im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
#im_sized = random_flip(image, flip)
flip2 = np.random.randint(2)
im_sized = random_flip2(im_sized, flip2)
# correct the size and pos of bounding boxes
#all_objs = correct_bounding_boxes(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, image_w, image_h)
all_objs = correct_bounding_boxes2(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, flip2, image_w, image_h)
return im_sized, all_objs
def _aug_image(self, instance, net_h, net_w):
image_name = instance['filename']
image = cv2.imread(image_name, ) # RGB image
image = cv2.imdecode(np.fromfile(image_name, dtype=np.uint8), cv2.IMREAD_UNCHANGED) # 打开含有中文路径的图片
if image is None: print('Cannot find ', image_name)
image = image[:,:,::-1] # RGB image
image_h, image_w, _ = image.shape
# determine the amount of scaling and cropping
dw = self.jitter * image_w;
dh = self.jitter * image_h;
new_ar = (image_w + np.random.uniform(-dw, dw)) / (image_h + np.random.uniform(-dh, dh));
scale = np.random.uniform(0.25, 2);
if (new_ar < 1):
new_h = int(scale * net_h);
new_w = int(net_h * new_ar);
else:
new_w = int(scale * net_w);
new_h = int(net_w / new_ar);
dx = int(np.random.uniform(0, net_w - new_w));
dy = int(np.random.uniform(0, net_h - new_h));
# apply scaling and cropping
im_sized = apply_random_scale_and_crop(image, new_w, new_h, net_w, net_h, dx, dy)
# randomly distort hsv space
im_sized = random_distort_image(im_sized)
# randomly flip
flip = np.random.randint(2)
im_sized = random_flip(im_sized, flip)
# correct the size and pos of bounding boxes
all_objs = correct_bounding_boxes(instance['object'], new_w, new_h, net_w, net_h, dx, dy, flip, image_w, image_h)
return im_sized, all_objs
