def get_augumentation(phase, width=512, height=512, min_area=0., min_visibility=0., ft='coco'):
# from albumentations for all detection and segmentation
list_transforms = []
if phase == 'train':
list_transforms.extend([
transforms.LongestMaxSize(
max_size=width, always_apply=True),
albu.PadIfNeeded(min_height=height, min_width=width,
always_apply=True, border_mode=0, value=[0, 0, 0]),
transforms.RandomResizedCrop(
height=height,
width=width, p=0.3),
transforms.Flip(),
transforms.Transpose(),
albu.OneOf([
albu.RandomBrightnessContrast(brightness_limit=0.5,
contrast_limit=0.4),
albu.RandomGamma(gamma_limit=(50, 150)),
albu.NoOp()
]),
albu.OneOf([
albu.RGBShift(r_shift_limit=20, b_shift_limit=15,
g_shift_limit=15),
albu.HueSaturationValue(hue_shift_limit=5,
sat_shift_limit=5),
albu.NoOp()
]),
albu.CLAHE(p=0.8),
albu.HorizontalFlip(p=0.5),
albu.VerticalFlip(p=0.5),
])
if phase == 'test':
list_transforms.extend([
albu.Resize(height=height, width=width)
])
list_transforms.extend([
albu.Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225), p=1),
ToTensorV2()
])
if phase == 'test':
return albu.Compose(list_transforms)
return albu.Compose(list_transforms, bbox_params=albu.BboxParams(format=ft,
min_area=min_area,
min_visibility=min_visibility,
label_fields=['category_id']))
def get_augmentations():
"""Get a list of 'major' and 'minor' augmentation functions for the pipeline in a dictionary."""
return {
"major": {
"shift-scale-rot":
trans.ShiftScaleRotate(
shift_limit=0.05,
rotate_limit=35,
border_mode=cv2.BORDER_REPLICATE,
always_apply=True,
),
"crop":
trans.RandomResizedCrop(100,
100,
scale=(0.8, 0.95),
ratio=(0.8, 1.2),
always_apply=True),
# "elastic": trans.ElasticTransform(
# alpha=0.8,
# alpha_affine=10,
# sigma=40,
# border_mode=cv2.BORDER_REPLICATE,
# always_apply=True,
# ),
"distort":
trans.OpticalDistortion(0.2, always_apply=True),
},
"minor": {
"blur":
trans.GaussianBlur(7, always_apply=True),
"noise":
trans.GaussNoise((20.0, 40.0), always_apply=True),
"bright-contrast":
trans.RandomBrightnessContrast(0.4, 0.4, always_apply=True),
"hsv":
trans.HueSaturationValue(30, 40, 50, always_apply=True),
"rgb":
trans.RGBShift(always_apply=True),
"flip":
trans.HorizontalFlip(always_apply=True),
},
}
def train_dataloader(self):
augmentations = Compose(
[
A.RandomResizedCrop(
height=self.hparams.sz,
width=self.hparams.sz,
scale=(0.7, 1.0),
),
# AdvancedHairAugmentation(),
A.GridDistortion(),
A.RandomBrightnessContrast(),
A.ShiftScaleRotate(),
A.Flip(p=0.5),
A.CoarseDropout(
max_height=int(self.hparams.sz / 10),
max_width=int(self.hparams.sz / 10),
),
# A.HueSaturationValue(),
A.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225],
max_pixel_value=255,
),
ToTensorV2(),
]
)
train_ds = MelanomaDataset(
df=self.train_df,
images_path=self.train_images_path,
augmentations=augmentations,
train_or_valid=True,
)
return DataLoader(
train_ds,
# sampler=sampler,
batch_size=self.hparams.bs,
shuffle=True,
num_workers=os.cpu_count(),
pin_memory=True,
)
def get_tta_transforms():
return Compose([
A.RandomResizedCrop(
height=hparams.sz,
width=hparams.sz,
scale=(0.7, 1.0),
),
# AdvancedHairAugmentation(),
A.GridDistortion(),
A.RandomBrightnessContrast(),
A.ShiftScaleRotate(),
A.Flip(p=0.5),
A.CoarseDropout(
max_height=int(hparams.sz / 10),
max_width=int(hparams.sz / 10),
),
# A.HueSaturationValue(),
A.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225],
max_pixel_value=255,
),
ToTensorV2(),
])