def get_train_transform(image_size=our_image_size):
return A.Compose(
[
A.RandomResizedCrop(height=image_size,
width=image_size,
scale=(0.5, 1.5),
ratio=(0.75, 1.25),
p=1),
A.Resize(height=image_size, width=image_size, p=1.0),
# Add occasion blur
A.OneOf([A.GaussianBlur(), A.MotionBlur()], p=0.5),
A.Rotate(limit=90, border_mode=cv2.BORDER_CONSTANT, value=0,
p=0.5),
# noise
A.OneOf([
A.GaussNoise(p=0.5),
A.RandomGamma(p=0.4),
], p=0.5),
# D4 Augmentations
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
A.Transpose(p=0.2),
# Spatial-preserving augmentations
A.HueSaturationValue(hue_shift_limit=0.2,
sat_shift_limit=0.2,
val_shift_limit=0.2,
p=0.7),
A.RandomBrightnessContrast(
brightness_limit=0.2, contrast_limit=0.2, p=0.8),
A.ToGray(p=0.01),
# cutout
A.Cutout(num_holes=32,
max_h_size=image_size // 16,
max_w_size=image_size // 16,
fill_value=0,
p=0.5),
],
bbox_params={
'format': 'pascal_voc',
'label_fields': ['labels']
})
def mix_transform(resize):
return A.Compose([
pre_transform(resize=resize),
A.GridDistortion(),
A.RandomRotate90(always_apply=True),
A.Rotate(limit=10, border_mode=0, p=0.5),
# A.OneOf([
A.Flip(p=0.5),
A.Transpose(p=0.5),
# ], p=0.5),
A.RandomBrightnessContrast(0.2, 0.2, p=0.3),
A.OneOf([
A.Blur(blur_limit=3, p=0.2),
A.MotionBlur(blur_limit=5, p=0.4),
A.MedianBlur(blur_limit=5, p=0.4)
],
p=0.3),
A.GaussNoise((10, 50), p=0.1),
post_transform()
])
def __init__(self, crop_size=1024):
self.master = albumentations.Compose([
albumentations.RandomCrop(crop_size, crop_size),
albumentations.RandomRotate90(p=0.5),
albumentations.Transpose(p=0.5),
albumentations.Flip(p=0.5),
albumentations.OneOf([
albumentations.RandomBrightness(),
albumentations.RandomContrast(),
albumentations.HueSaturationValue(),
],
p=0.5),
albumentations.ElasticTransform(),
albumentations.ShiftScaleRotate(shift_limit=0.1,
scale_limit=0.1,
rotate_limit=15,
p=0.5),
albumentations.Normalize(mean=[0.798, 0.621, 0.841],
std=[0.125, 0.228, 0.089]),
])
def _image_augment(self, img): #, bboxes):
'''
image_augment
ARGS: img - numpy array of image
bboxes - array of bounding boxes in COCO format
'''
transform = A.Compose([
A.HorizontalFlip(p=0.5),
A.HorizontalFlip(p=0.5),
A.RandomRotate90(p=0.5),
A.Rotate(p=0.5),
A.RandomBrightnessContrast(p=0.3),
]) #, bbox_params=A.BboxParams(format='coco'))
transformed = transform(image=img,
# bboxes=bboxes
)
transformed_image = transformed['image']
# transformed_bboxes = transformed['bboxes']
return transformed_image #, transformed_bboxes
def test_transform_pipeline_serialization_with_keypoints(seed, image, keypoints, keypoint_format, labels):
aug = A.Compose(
[
A.OneOrOther(
A.Compose([A.RandomRotate90(), A.OneOf([A.HorizontalFlip(p=0.5), A.VerticalFlip(p=0.5)])]),
A.Compose([A.Rotate(p=0.5), A.OneOf([A.HueSaturationValue(p=0.5), A.RGBShift(p=0.7)], p=1)]),
),
A.HorizontalFlip(p=1),
A.RandomBrightnessContrast(p=0.5),
],
keypoint_params={"format": keypoint_format, "label_fields": ["labels"]},
)
serialized_aug = A.to_dict(aug)
deserialized_aug = A.from_dict(serialized_aug)
set_seed(seed)
aug_data = aug(image=image, keypoints=keypoints, labels=labels)
set_seed(seed)
deserialized_aug_data = deserialized_aug(image=image, keypoints=keypoints, labels=labels)
assert np.array_equal(aug_data["image"], deserialized_aug_data["image"])
assert np.array_equal(aug_data["keypoints"], deserialized_aug_data["keypoints"])
def create_train_transforms(self, crop_size):
transforms = [
A.OneOf([
RandomCrop(crop_size, crop_size, p=0.3),
RandomSizedCropAroundBbox(min_max_height=(int(
crop_size * 0.65), int(crop_size * 1.4)),
height=crop_size,
width=crop_size,
p=0.7)
],
p=1),
A.Rotate(20, p=0.2, border_mode=cv2.BORDER_CONSTANT),
A.HorizontalFlip(),
A.VerticalFlip(),
A.RandomRotate90(),
A.RandomBrightnessContrast(),
A.RandomGamma(),
A.FancyPCA(p=0.2)
]
return A.Compose(transforms, additional_targets={'labels': 'mask'})
def hard_augmentations():
return A.Compose([
A.RandomRotate90(),
A.Transpose(),
A.RandomGridShuffle(),
A.ShiftScaleRotate(scale_limit=0.1,
rotate_limit=45,
border_mode=cv2.BORDER_CONSTANT,
mask_value=0,
value=0),
A.ElasticTransform(border_mode=cv2.BORDER_CONSTANT,
alpha_affine=5,
mask_value=0,
value=0),
# Add occasion blur
A.OneOf([
A.GaussianBlur(),
A.GaussNoise(),
A.IAAAdditiveGaussianNoise(),
A.NoOp()
]),
# D4 Augmentations
A.OneOf([A.CoarseDropout(),
A.MaskDropout(max_objects=10),
A.NoOp()]),
# Spatial-preserving augmentations:
A.OneOf([
A.RandomBrightnessContrast(brightness_by_max=True),
A.CLAHE(),
A.HueSaturationValue(),
A.RGBShift(),
A.RandomGamma(),
A.NoOp(),
]),
# Weather effects
A.OneOf([
A.RandomFog(fog_coef_lower=0.01, fog_coef_upper=0.3, p=0.1),
A.NoOp()
]),
A.Normalize(),
])
def train_transform(width=512, height=512, min_area=0.0, min_visibility=0.0, lamda_norm=False):
list_transforms = []
augment = albu.Compose([
albu.OneOf(
[
albu.RandomSizedBBoxSafeCrop(p=1.0, height=height, width=width),
albu.HorizontalFlip(p=1.0),
albu.VerticalFlip(p=1.0),
albu.RandomRotate90(p=1.0),
albu.NoOp(p=1.0)
]
),
albu.OneOf(
[
albu.RandomBrightnessContrast(p=1.0),
albu.RandomGamma(p=1.0),
albu.NoOp(p=1.0)
]
),
albu.OneOf(
[
albu.MotionBlur(p=1.0),
albu.RandomFog(p=1.0),
albu.RandomRain(p=1.0),
albu.CLAHE(p=1.0),
albu.ToGray(p=1.0),
albu.NoOp(p=1.0)
]
)
])
list_transforms.extend([augment])
if lamda_norm:
list_transforms.extend([albu.Lambda(image=lamda_norm_tran)])
else:
list_transforms.extend([albu.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225), max_pixel_value=255., p=1.0)])
list_transforms.extend([albu.Resize(height=height, width=width, p=1.0)])
return albu.Compose(list_transforms, bbox_params=albu.BboxParams(format='pascal_voc', min_area=min_area,
min_visibility=min_visibility,
label_fields=['label']))
def __init__(self, img, data, img_size):
"""
arguments
---------
img : list
list of images, in the original size (height, width, 3)
data : list of dict
Each dict has :
'image' : index of the image. The index should match with img
'mask' : [xx, yy]
IMPORTANT : (WIDTH, HEIGHT)
'box' : [[xmin, ymin], [xmax,ymax]]
'size' : the size of the image that the data was created with
IMPORTANT : (WIDTH, HEIGHT)
img_size : tuple
Desired output image size
The axes will be swapped to match pygame.
IMPORTANT : (WIDTH, HEIGHT)
"""
self.image = img
self.data = data
self.n = len(data)
self.output_size = img_size
self.aug = A.Compose([
A.OneOf([
A.RandomGamma((40, 200), p=1),
A.RandomBrightness(limit=0.5, p=1),
A.RandomContrast(limit=0.5, p=1),
A.RGBShift(40, 40, 40, p=1),
A.Downscale(scale_min=0.25, scale_max=0.5, p=1),
A.ChannelShuffle(p=1),
],
p=0.8),
A.InvertImg(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=1),
A.Resize(img_size[0], img_size[1]),
], )
for datum in data:
datum['mask_min'] = np.min(datum['mask'], axis=1)
datum['mask_max'] = np.max(datum['mask'], axis=1) + 1
def get_train_transform():
train_pipline = [
PhotoMetricDistortion(
brightness_delta=32,
contrast_range=(0.5, 1.5),
saturation_range=(0.5, 1.5),
hue_delta=18),
MixUp(p=0.5, mode=1),
# Mosaic(p=0.2),
GaussNoise(p=0.2),
A.Compose([
A.Flip(p=0.5),
A.RandomRotate90(p=0.5),
A.ToGray(p=0.01),
A.Cutout(num_holes=8, max_h_size=64, max_w_size=64, fill_value=0, p=0.5),
A.RandomCrop(height=1000, width=1000, p=0.5),
ToTensorV2(p=1.0)
], bbox_params={'format': 'pascal_voc', 'label_fields': ['labels']})
]
return train_pipline
def get_validation_augmentation(self, conf, is_stub=False):
# Since batch-size in validation is 1, validation could be performed by whole crop-size.
# To provide pos
test_transform = [
alb.HorizontalFlip(p=0.5),
alb.VerticalFlip(p=0.5),
alb.RandomRotate90(always_apply=False, p=0.5),
alb.PadIfNeeded(conf.img_wh_crop,
conf.img_wh_crop,
always_apply=True,
border_mode=0),
# alb.RandomCrop(height=conf.img_wh_crop, width=conf.img_wh_crop, always_apply=True),
]
if is_stub:
return alb.Compose([
alb.PadIfNeeded(conf.img_wh_crop,
conf.img_wh_crop,
always_apply=True,
border_mode=0)
])
return alb.Compose(test_transform)
def generate_ds(size):
trfm = A.Compose([
A.Resize(size, size, p=1.0),
A.HorizontalFlip(),
A.VerticalFlip(),
A.RandomRotate90(),
A.ShiftScaleRotate(shift_limit=0.0625, scale_limit=0.2, rotate_limit=20, p=0.9,
border_mode=cv2.BORDER_REFLECT),
A.OneOf([
A.OpticalDistortion(p=0.4),
A.GridDistortion(p=.1),
A.IAAPiecewiseAffine(p=0.4),
], p=0.3),
A.OneOf([
A.HueSaturationValue(10,15,10),
A.CLAHE(clip_limit=3),
A.RandomBrightnessContrast(),
], p=0.5)
], p=1.0)
return HubDataset(DATA_PATH, window=WINDOW, overlap=MIN_OVERLAP, transform=trfm)
def get_training_augmentation():
train_transform = [
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=1.),
# A.ShiftScaleRotate(scale_limit=0.5, rotate_limit=0, shift_limit=0.1, p=1., border_mode=0),
#
# A.PadIfNeeded(min_height=128, min_width=128, always_apply=True, border_mode=0),
# A.RandomCrop(height=128, width=128, always_apply=True),
#
# A.IAAAdditiveGaussianNoise(p=0.2),
# A.IAAPerspective(p=0.5),
#
# A.OneOf(
# [
# A.CLAHE(p=1.),
# A.RandomBrightness(p=1.),
# A.RandomGamma(p=1.),
# ],
# p=0.9,
# ),
#
# A.OneOf(
# [
# A.IAASharpen(p=1.),
# A.Blur(blur_limit=3., p=1.),
# A.MotionBlur(blur_limit=3., p=1.),
# ],
# p=0.9,
# ),
#
# A.OneOf(
# [
# A.RandomContrast(p=1.),
# A.HueSaturationValue(p=1.),
# ],
# p=0.9,
# ),
]
return A.Compose(train_transform)
def data_augmentation(image_path, aug_num):
image = cv2.imread(image_path, cv2.IMREAD_COLOR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
augmentation = A.Compose(
[
A.RandomRotate90(),
A.Flip(),
A.Transpose(),
A.OneOf([
A.IAAAdditiveGaussianNoise(),
A.GaussNoise(),
], p=0.2),
A.OneOf([
A.MotionBlur(p=0.2),
A.MedianBlur(blur_limit=3, p=0.1),
A.Blur(blur_limit=3, p=0.1),
],
p=0.2),
A.ShiftScaleRotate(
shift_limit=0.0625, scale_limit=0.2, rotate_limit=45, p=0.2),
A.OneOf([
A.OpticalDistortion(p=0.3),
A.GridDistortion(p=0.1),
A.IAAPiecewiseAffine(p=0.3),
],
p=0.2),
A.OneOf([
A.CLAHE(clip_limit=2),
A.IAASharpen(),
A.IAAEmboss(),
A.RandomBrightnessContrast(),
],
p=0.3),
# A.HueSaturationValue(p=0.3),
],
p=0.5)
patches = []
for _ in range(aug_num):
patches.append(augmentation(image=image)['image'])
return patches
def affine_transform(
self): # Affine Transforms: Scale & Translation & Rotation
aug = A.Compose(
[
A.Transpose(
p=self.p
), # Transpose the input by swapping rows and columns.
A.OneOf(
[
A.RandomRotate90(
p=self.p
), # Randomly rotate the input by 90 degrees zero or more times.
A.Rotate(
limit=90, p=self.p
), # Rotate the input by an angle selected randomly from the uniform distribution.
A.ShiftScaleRotate(
shift_limit=0.0625,
scale_limit=0.1,
rotate_limit=45,
p=self.p
) # Randomly apply affine transforms: translate, scale and rotate the input.
],
p=1),
A.OneOf(
[
A.HorizontalFlip(
p=self.p
), # Flip the input vertically around the x-axis.
A.VerticalFlip(
p=self.p
), # Flip the input horizontally around the y-axis.
A.Flip(
p=self.p
) # Flip the input either horizontally, vertically or both horizontally and vertically.
],
p=1)
],
p=1)
return aug
def train(args):
images, y = data.load(args.data_dir)
X = np.array(
[model.crop_nail(image, height=320, width=320) for image in images])
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=args.test_size, random_state=0)
net = model.create()
X_test = np.array(
[model.resize(image, net.input_shape[1:3]) for image in X_test])
X_test = model.preprocess_image(X_test)
augmentator = albumentations.Compose([
albumentations.ShiftScaleRotate(shift_limit=0.1,
scale_limit=0.1,
rotate_limit=45,
p=.8),
albumentations.OneOf([
albumentations.RandomRotate90(),
albumentations.Flip(),
],
p=.6),
albumentations.Resize(width=net.input_shape[2],
height=net.input_shape[1]),
albumentations.GaussNoise(var_limit=(1, 3)),
])
net.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
net.fit_generator(
AugmentedGenerator(X_train, y_train, 40, augmentator),
epochs=args.epochs,
validation_data=(X_test, y_test),
)
net.save_weights(args.target, overwrite=True)
def stong_aug():
# strong aug for train
train_transform = [
albu.Resize(height=configs.input_size, width=configs.input_size),
albu.HorizontalFlip(p=0.5),
albu.VerticalFlip(p=0.5),
albu.RandomRotate90(p=0.5),
albu.OneOf([
albu.CenterCrop(
p=0.5, height=configs.input_size, width=configs.input_size),
albu.ElasticTransform(
p=0.5, alpha=120, sigma=120 * 0.05, alpha_affine=120 * 0.03),
albu.GridDistortion(p=0.5),
albu.OpticalDistortion(p=1, distort_limit=1, shift_limit=0.5),
],
p=0.8),
albu.Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225),
p=1),
AT.ToTensor(),
]
return albu.Compose(train_transform)
def get_valid_transforms():
return A.Compose([
# A.ShiftScaleRotate(p=0.5),
A.RandomRotate90(p=0.5),
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
# A.RandomBrightness(p=0.5),
# A.RandomContrast(p=0.5),
# A.RandomGamma(p=0.5),
# A.RGBShift(),
# A.CLAHE(p=1),
# A.ToGray(),
# A.OneOf([
# A.RandomContrast(),
# A.RandomGamma(),
# A.RandomBrightness(),
# ], p=0.5),
# A.HueSaturationValue(p=0.5),
# A.ChannelShuffle(p=0.5),
#A.Resize(height=384, width=384, interpolation=cv2.INTER_AREA, p=1),
ToTensorV2(p=1.0),
], p=1.0)
def get_transform(image, mask, original_h, original_w):
'''
'''
transform = A.Compose([
A.OneOf([
A.VerticalFlip(p=0.2),
A.RandomRotate90(p=0.2),
A.HorizontalFlip(p=0.2)
],
p=0.2),
A.OneOf([
A.MotionBlur(p=0.2),
A.MedianBlur(blur_limit=15, p=0.2),
A.Blur(blur_limit=15, p=0.2),
A.GaussNoise(p=0.2),
],
p=0.2),
])
return transform
def resize_transforms(input_size=256):
result = A.Compose([
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=1),
A.OneOf([
A.ShiftScaleRotate(scale_limit=0,
rotate_limit=(-45, 45),
shift_limit=(-0.1, 0.1),
interpolation=0,
border_mode=2,
p=0.5),
A.ElasticTransform(alpha_affine=20, sigma=30, border_mode=2, p=0.5)
]),
A.PadIfNeeded(min_height=input_size,
min_width=input_size,
always_apply=True,
border_mode=2),
A.RandomCrop(input_size, input_size, always_apply=True),
])
return result
def __init__(self,
input_key: str = "image",
output_key: str = "rotation_factor",
targets_key: str = None,
rotate_probability: float = 1.,
hflip_probability: float = 0.5,
one_hot_classes: int = None):
"""
Args:
input_key (str): input key to use from annotation dict
output_key (str): output key to use to store the result
"""
self.input_key = input_key
self.output_key = output_key
self.targets_key = targets_key
self.rotate_probability = rotate_probability
self.hflip_probability = hflip_probability
self.rotate = A.RandomRotate90()
self.hflip = A.HorizontalFlip()
self.one_hot_classes = one_hot_classes * 8 \
if one_hot_classes is not None \
else None
def __init__(self, img, data, img_size):
"""
arguments
---------
img : list
list of images, in the original size (height, width, 3)
data : list of dict
Each dict has :
'image' : index of the image. The index should match with img
'mask' : [rr, cc]
'box' : [[xmin, ymin], [xmax,ymax]]
'size' : the size of the image that the data was created with
IMPORTANT : (WIDTH, HEIGHT)
img_size : tuple
Desired output image size
The axes will be swapped to match pygame.
IMPORTANT : (WIDTH, HEIGHT)
"""
self.img = img
self.data = data
self.n = len(data)
self.output_size = img_size
self.aug = A.Compose(
[
A.OneOf([
A.RandomGamma((40, 200), p=1),
A.RandomBrightness(limit=0.5, p=1),
A.RandomContrast(limit=0.5, p=1),
A.RGBShift(p=1),
],
p=0.8),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=1),
A.Resize(img_size[1], img_size[0]),
],
bbox_params=A.BboxParams(format='albumentations',
label_fields=['bbox_classes']),
keypoint_params=A.KeypointParams(format='xy'),
)
def random_rotate_90(image: np.ndarray,
annotations: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
image_height, image_width = image.shape[:2]
boxes = annotations[:, :4]
keypoints = annotations[:, 4:-1].reshape(-1, 2)
labels = annotations[:, -1:]
invalid_index = keypoints.sum(axis=1) == -2
keypoints[:, 0] = np.clip(keypoints[:, 0], 0, image_width - 1)
keypoints[:, 1] = np.clip(keypoints[:, 1], 0, image_height - 1)
keypoints[invalid_index] = 0
category_ids = list(range(boxes.shape[0]))
transform = albu.Compose(
[albu.RandomRotate90(p=1)],
keypoint_params=albu.KeypointParams(format="xy"),
bbox_params=albu.BboxParams(format="pascal_voc",
label_fields=["category_ids"]),
)
transformed = transform(image=image,
keypoints=keypoints.tolist(),
bboxes=boxes.tolist(),
category_ids=category_ids)
keypoints = np.array(transformed["keypoints"])
keypoints[invalid_index] = -1
keypoints = keypoints.reshape(-1, 10)
boxes = np.array(transformed["bboxes"])
image = transformed["image"]
annotations = np.hstack([boxes, keypoints, labels])
return image, annotations
def clf_train_augs(height: int, width: int) -> albu.Compose:
"""Training augmentations for classification. We prefer for this model to be really
robust. Feel free to tweak these paramters or ad other augmentations."""
return albu.Compose([
albu.Resize(height=height, width=width),
albu.OneOf([
albu.IAAAffine(shear=6, rotate=5, always_apply=True),
albu.ShiftScaleRotate(shift_limit=0.025,
scale_limit=0.1,
rotate_limit=10),
]),
albu.ShiftScaleRotate(shift_limit=0.025,
scale_limit=0.1,
rotate_limit=10),
albu.Flip(),
albu.RandomRotate90(),
albu.OneOf(
[
albu.HueSaturationValue(p=1.0),
albu.IAAAdditiveGaussianNoise(p=1.0),
albu.IAASharpen(p=1.0),
albu.RandomBrightnessContrast(
brightness_limit=0.1, contrast_limit=0.1, p=1.0),
albu.RandomGamma(p=1.0),
],
p=1.0,
),
albu.OneOf(
[
albu.Blur(blur_limit=3, p=1.0),
albu.MedianBlur(blur_limit=3, p=1.0),
albu.MotionBlur(blur_limit=3, p=1.0),
],
p=1.0,
),
albu.Normalize(),
])
def get_albumentations_train_transforms():
return A.Compose(
[
A.RandomSizedCrop(
min_max_height=(600, 800), height=1024, width=1024, p=0.5),
A.OneOf([
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
p=1.0),
A.RandomBrightnessContrast(
brightness_limit=0.2, contrast_limit=0.2, p=1.0),
],
p=0.8),
A.OneOf(
[
A.RGBShift(p=1.0),
A.CLAHE(p=1.0), # internal logic is rgb order
A.RandomGamma(p=1.0),
],
p=0.4),
# A.CLAHE(p=0.3),
A.ToGray(p=0.01),
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
A.Cutout(num_holes=8,
max_h_size=64,
max_w_size=64,
fill_value=(124, 117, 104),
p=0.5), # rgb order
],
p=1.0,
bbox_params=A.BboxParams(format='coco',
min_area=4,
min_visibility=0.01,
label_fields=['category_id']))
def __init__(self):
self.val_transform = A.Compose([
A.Resize(224, 224),
])
self.test_transform = A.Compose([
A.Resize(224, 224),
])
self.train_transform = A.Compose([
A.Resize(224, 224),
A.HueSaturationValue(),
A.RandomRotate90(),
A.ShiftScaleRotate(rotate_limit=15),
A.Blur(),
A.OpticalDistortion(),
# A.GridDistortion(),
A.Cutout(num_holes=16, max_h_size=16, max_w_size=16),
])
# split ratio between train and val dataset
ratio = 0.8
# Data file list
self.train_list = pd.read_csv(f"{DATASET}/train_list.txt",
index_col=False,
sep="\t").to_numpy()
size = self.train_list.shape[0]
self.offset = int(size * ratio)
np.random.shuffle(self.train_list)
logger.info("after shuffle: {}".format(self.train_list[:, 1]))
test_list_temp = glob(f"{DATASET}/test/*.jpg")
self.test_list = []
for path in test_list_temp:
self.test_list.append(path[len(DATASET) + 1:])
def hard_augmentations(mask_dropout=True) -> List[A.DualTransform]:
return [
# D4 Augmentations
A.RandomRotate90(p=1),
A.Transpose(p=0.5),
# Spatial augmentations
A.OneOf(
[
A.ShiftScaleRotate(scale_limit=0.2, rotate_limit=45, border_mode=cv2.BORDER_REFLECT101),
A.ElasticTransform(border_mode=cv2.BORDER_REFLECT101, alpha_affine=5),
]
),
# Color augmentations
A.OneOf(
[
A.RandomBrightnessContrast(brightness_by_max=True),
A.CLAHE(),
A.FancyPCA(),
A.HueSaturationValue(),
A.RGBShift(),
A.RandomGamma(),
]
),
# Dropout & Shuffle
A.OneOf(
[
A.RandomGridShuffle(),
A.CoarseDropout(),
A.MaskDropout(max_objects=2, mask_fill_value=0) if mask_dropout else A.NoOp(),
]
),
# Add occasion blur
A.OneOf([A.GaussianBlur(), A.GaussNoise(), A.IAAAdditiveGaussianNoise()]),
# Weather effects
A.RandomFog(fog_coef_lower=0.01, fog_coef_upper=0.3, p=0.1),
]
def heavy_training_transforms():
return A.Compose([
A.OneOf([
A.Transpose(),
A.VerticalFlip(),
A.HorizontalFlip(),
A.RandomRotate90(),
A.NoOp()
],
p=1.0),
A.OneOf([
A.CLAHE(),
A.RGBShift(),
A.RandomBrightnessContrast(),
A.RandomGamma(),
A.HueSaturationValue(),
A.NoOp()
],
p=0.15),
A.OneOf([
A.ElasticTransform(),
A.GridDistortion(),
A.OpticalDistortion(),
A.NoOp(),
A.ShiftScaleRotate(),
],
p=1.0),
A.OneOf([A.GaussNoise(), A.GaussianBlur(),
A.NoOp()], p=0.15),
A.OneOf([
A.CoarseDropout(max_holes=16, max_height=16, max_width=16),
A.NoOp()
],
p=1.0),
A.Normalize()
])
def get_augmentation_transforms(add_non_spatial=True):
transforms = [
A.VerticalFlip(p=0.5),
A.HorizontalFlip(p=0.5),
A.RandomRotate90(p=0.5),
A.OneOf([
A.ElasticTransform(
alpha=120, sigma=120 * 0.05, alpha_affine=120 * 0.03, p=0.5),
A.GridDistortion(p=0.5),
A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=1)
],
p=0.8),
]
if add_non_spatial:
transforms.extend([
A.CLAHE(p=0.8),
A.RandomBrightnessContrast(p=0.8),
A.RandomGamma(p=0.8)
])
aug_transform = A.Compose(transforms)
return aug_transform
def __init__(self, data_dir, train=True, ptsize=128):
#
super(F3DS, self).__init__()
self.train = train
self.ptsize = ptsize
self.data_dir = data_dir
self.get_data() #
self.mapDataTo1() # 先归一化到[0, 1]之间，并用均值标准化，
self.hw = self.data_cube.shape[1:] # 图片大小 [463+20, 951+20,]
self.train_seg = [0, 1, 3]
self.test_seg = [2]
if train==True: # 训练数据和测试数据
self.data_cube = self.data_cube[self.train_seg]
self.label_cube = self.label_cube[self.train_seg]
else: # test
self.data_cube = self.data_cube[self.test_seg]
self.label_cube = self.label_cube[self.test_seg]
c = 20
if self.train:
self.hwarange = [torch.arange(0, x-ptsize+c, 10) for x in self.hw]
else:
self.hwarange = [torch.arange(0, x-ptsize, ptsize) for x in self.hw]
self.chwn = [self.data_cube.shape[0] ] + [len(x) for x in self.hwarange] # 切面数、h方向滑动个数、w方向滑动个数
self.augment = A.Compose([
A.PadIfNeeded(min_height=self.ptsize+c, min_width=self.ptsize+c, ),
A.RandomRotate90(),
A.HorizontalFlip(),
A.VerticalFlip(),
])
self.tsf = transforms.Compose([
transforms.ToTensor(),
])