def __init__(self, n, m):
self.n = n
self.m = m
m_ratio = self.m / 30.0
self.augment_list = (
A.CLAHE(always_apply=True),
A.Equalize(always_apply=True),
A.InvertImg(always_apply=True),
A.Rotate(limit=30 * m_ratio, always_apply=True),
A.Posterize(num_bits=int(4 * m_ratio), always_apply=True),
A.Solarize(threshold=m_ratio, always_apply=True),
A.RGBShift(r_shift_limit=110 * m_ratio,
g_shift_limit=110 * m_ratio,
b_shift_limit=110 * m_ratio,
always_apply=True),
A.HueSaturationValue(hue_shift_limit=20 * m_ratio,
sat_shift_limit=30 * m_ratio,
val_shift_limit=20 * m_ratio,
always_apply=True),
A.RandomContrast(limit=m_ratio, always_apply=True),
A.RandomBrightness(limit=m_ratio, always_apply=True),
# A.Sharpen(always_apply=True), 0.1, 1.9),
A.ShiftScaleRotate(shift_limit=0.3 * m_ratio,
shift_limit_y=0,
rotate_limit=0,
always_apply=True),
A.ShiftScaleRotate(shift_limit=0.3 * m_ratio,
shift_limit_x=0,
rotate_limit=0,
always_apply=True),
A.Cutout(num_holes=int(8 * m_ratio), always_apply=True),
A.IAAAffine(shear=0.3 * m_ratio, always_apply=True))
assert self.n <= len(self.augment_list)
def __init__(self,fold_file, pkl_file_path, folds, image_height, image_width, mean, std):
self.pkl_file_path = pkl_file_path
self.fold_file = fold_file
df = pd.read_csv(self.fold_file)
df = df[['image_id','labels','kfold']]
df = df[df['kfold'].isin(folds)].reset_index(drop= True)
class_map = {'A':0,'B':1,'C':2}
self.img_id = df['image_id'].apply(lambda x: x.split('.')[0]).values
self.labels = df['labels'].apply(lambda x: x[-1]).map(class_map).values
# training set
self.aug = albumentations.Compose([
albumentations.Resize(image_height, image_width, always_apply=True),
albumentations.Cutout(num_holes=8,max_h_size=30,max_w_size=60,always_apply=False,p=0.9),
albumentations.Equalize(always_apply=False, p=0.9),
albumentations.RandomShadow(shadow_roi=(0, 0.85, 1, 1), p=0.9),
albumentations.RandomBrightnessContrast(brightness_limit=0.1, contrast_limit=0.1, p=0.9),
albumentations.ShiftScaleRotate(shift_limit=0.0625,
scale_limit=0.1,
rotate_limit=5,
p=0.9),
albumentations.Normalize(mean, std, always_apply=True)
])
def test_equalize():
aug = A.Equalize(p=1)
img = np.random.randint(0, 256, 256 * 256 * 3, np.uint8).reshape((256, 256, 3))
a = aug(image=img)["image"]
b = F.equalize(img)
assert np.all(a == b)
mask = np.random.randint(0, 2, 256 * 256, np.uint8).reshape((256, 256))
aug = A.Equalize(mask=mask, p=1)
a = aug(image=img)["image"]
b = F.equalize(img, mask=mask)
assert np.all(a == b)
def mask_func(image, test):
return mask
aug = A.Equalize(mask=mask_func, mask_params=["test"], p=1)
assert np.all(aug(image=img, test=mask)["image"] == F.equalize(img, mask=mask))
def augment_and_save(path_to_get_data,
path_to_save_data,
number_of_tranformation=10):
"""
Function defined to apply an image / rounding boxes transformation pipeline
and save the corresponding files.
Args:
-----
- path_to_get_data: str, the folder path where untouched data is
- path_to_save_data: str, the folder path where to save augmented data
- number_of_tranformation: int, number of transformation to perform
Returns:
--------
- None
"""
images_names, yolo_names = get_images_and_box_files_names(path_to_get_data)
augmentation_pipeline = A.Compose([
A.Resize(416, 416),
A.Equalize(by_channels=True),
A.RGBShift(r_shift_limit=(-30, 30),
g_shift_limit=(-30, 30),
b_shift_limit=(-30, 30),
p=0.25),
A.HorizontalFlip(p=0.35),
A.VerticalFlip(p=0.35),
A.ShiftScaleRotate(border_mode=cv2.BORDER_REPLICATE, p=0.35),
A.RandomSnow(brightness_coeff=2.0, p=0.2)
], A.BboxParams('yolo', ['class_labels']))
# Iterate through each image
for idx, name in enumerate(images_names):
image_path = path_to_get_data + '/' + name
image = cv2.imread(image_path)
yolo_file_path = path_to_get_data + '/' + yolo_names[idx]
labels, coordinates = get_labels_and_coordinates(yolo_file_path)
# Generate x tranformation of the images
for i in tqdm(range(number_of_tranformation)):
new_image_name, new_yolos_name = set_new_files_names(
name, i, "jpg", "txt")
# Catch error due to unproper labelling
try:
new_image, new_coordinates, labels = get_data_from_pipeline(
augmentation_pipeline, image, coordinates, labels)
except ValueError as e:
print("**** Error Message ****\n")
print(f"{e}\n")
print(f"""Invalid transformation of box:
{str(new_coordinates)}\n""")
print(f"Image: {new_image_name}\n")
continue
# Same each image to jpg with its corresponding coordinates
save_image_bbox_data(path_to_save_data, new_image, new_image_name,
new_coordinates, labels, new_yolos_name)
def get_train_augm(size=Tuple[int, int],
p=0.5):
return albu.Compose([
albu.Resize(*size),
albu.OneOf([albu.CLAHE(6, (4, 4), always_apply=True),
albu.Equalize(always_apply=True)], p=0.99),
albu.HorizontalFlip(p=p),
albu.VerticalFlip(p=p),
albu.ToFloat(255),
ToTensor() # albu.Lambda(image=to_tensor)
])
def get_augmentation(save_path=None, load_path=None):
if load_path:
return A.load(load_path)
else:
aug_seq1 = A.OneOf([
A.Rotate(limit=(-90, 90), p=1.0),
A.Flip(p=1.0),
A.OpticalDistortion(always_apply=False, p=1.0, distort_limit=(-0.3, 0.3),
shift_limit=(-0.05, 0.05), interpolation=3,
border_mode=3, value=(0, 0, 0), mask_value=None),
], p=1.0)
aug_seq2 = A.OneOf([
# A.ChannelDropout(always_apply=False, p=1.0, channel_drop_range=(1, 1), fill_value=0),
A.RGBShift(r_shift_limit=15, g_shift_limit=15,
b_shift_limit=15, p=1.0),
A.RandomBrightnessContrast(always_apply=False, p=1.0, brightness_limit=(
-0.2, 0.2), contrast_limit=(-0.2, 0.2), brightness_by_max=True)
], p=1.0)
aug_seq3 = A.OneOf([
A.GaussNoise(always_apply=False, p=1.0, var_limit=(10, 50)),
A.ISONoise(always_apply=False, p=1.0, intensity=(
0.1, 1.0), color_shift=(0.01, 0.3)),
A.MultiplicativeNoise(always_apply=False, p=1.0, multiplier=(
0.8, 1.6), per_channel=True, elementwise=True),
], p=1.0)
aug_seq4 = A.OneOf([
A.Equalize(always_apply=False, p=1.0,
mode='pil', by_channels=True),
A.InvertImg(always_apply=False, p=1.0),
A.MotionBlur(always_apply=False, p=1.0, blur_limit=(3, 7)),
A.RandomFog(always_apply=False, p=1.0,
fog_coef_lower=0.01, fog_coef_upper=0.2, alpha_coef=0.2)
], p=1.0)
aug_seq = A.Compose([
# A.Resize(self.img_size, self.img_size),
# aug_seq1,
aug_seq2,
aug_seq3,
aug_seq4,
# A.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
# A.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]),
])
# aug_path = '/home/jitesh/prj/classification/test/bolt/aug/aug_seq.json'
if save_path:
A.save(aug_seq, save_path)
# loaded_transform = A.load(aug_path)
return aug_seq
def __getitem__(self, index):
"""Return a data point and its metadata information.
Parameters:
index - - a random integer for data indexing
Returns a dictionary that contains A, B, A_paths and B_paths
A (tensor) - - an image in the input domain
B (tensor) - - its corresponding image in the target domain
A_paths (str) - - image paths
B_paths (str) - - image paths (same as A_paths)
"""
# read a image given a random integer index
AB_path = self.AB_paths[index]
AB = Image.open(AB_path).convert('RGB')
# split AB image into A and B
w, h = AB.size
w2 = int(w / 2)
A = AB.crop((0, 0, w2, h))
B = AB.crop((w2, 0, w, h))
# apply the same transform to both A and B
# transform_params = get_params(self.opt, A.size)
transformOutput = Al.Compose([
Al.Crop(x_max=256, y_max=256),
Al.RandomRotate90(),
Al.Flip(),
])
transformInput = Al.Compose([
Al.Blur(blur_limit=3),
Al.Equalize(always_apply=True),
Al.RandomBrightnessContrast(),
transformOutput,
])
random.seed(42)
A = transformInput(image=np.array(A))['image']
B = transformOutput(image=np.array(B))['image']
# A_transform = get_transform(self.opt, transform_params, grayscale=(self.input_nc == 1))
# B_transform = get_transform(self.opt, transform_params, grayscale=(self.output_nc == 1))
# A = A_transform(A)
# B = B_transform(B)
return {'A': A, 'B': B, 'A_paths': AB_path, 'B_paths': AB_path}
def __init__(self, k: int = 5, always_apply: bool = True, p: float = 1.0):
super(RandAugmentAlb, self).__init__(always_apply, p)
self.k = k
self.candidates = [
AutoContrast(p=1.0),
A.Equalize(p=1.0),
A.InvertImg(p=1.0),
Rotate(30., p=1.0),
A.Posterize([4, 8], p=1.0),
A.Solarize([0, 256], p=1.0),
A.RandomBrightnessContrast(brightness_limit=0.,
contrast_limit=(0.05, 0.95),
p=1.0),
A.RandomBrightnessContrast(brightness_limit=(0.05, 0.95),
contrast_limit=0.,
p=1.0),
ShearX(0.3),
ShearY(0.3),
Translate(0.45),
]
def albumentations_list(MAGN: int = 4):
"""
Returns standard list of albumentations transforms, each of mangitude `MAGN`.
Args:
MAGN (int): Magnitude of each transform in the returned list.
"""
M = MAGN
transform_list = [
# PIXEL-LEVEL
A.RandomContrast(limit=M * .1, always_apply=True),
A.RandomBrightness(limit=M * .1, always_apply=True),
A.Equalize(always_apply=True),
A.OpticalDistortion(distort_limit=M * .2,
shift_limit=M * .1,
always_apply=True),
A.RGBShift(r_shift_limit=M * 10,
g_shift_limit=M * 10,
b_shift_limit=M * 10,
always_apply=True),
A.ISONoise(color_shift=(M * .01, M * .1),
intensity=(M * .02, M * .2),
always_apply=True),
A.RandomFog(fog_coef_lower=M * .01,
fog_coef_upper=M * .1,
always_apply=True),
A.CoarseDropout(max_holes=M * 10, always_apply=True),
A.GaussNoise(var_limit=(M, M * 50), always_apply=True),
# SPATIAL
A.Rotate(always_apply=True),
A.Transpose(always_apply=True),
A.NoOp(always_apply=True),
A.ElasticTransform(alpha=M * .25,
sigma=M * 3,
alpha_affine=M * 3,
always_apply=True),
A.GridDistortion(distort_limit=M * .075, always_apply=True)
]
return transform_list
def __init__(self, folds, img_height, img_width, mean, std):
df = pd.read_csv('../input/train_folds.csv')
df = df[[
'image_id', 'grapheme_root', 'vowel_diacritic',
'consonant_diacritic', 'kfold'
]]
df = df[df.kfold.isin(folds)].reset_index(drop=True)
self.image_ids = df.image_id.values
self.grapheme_root = df.grapheme_root.values
self.vowel_diacritic = df.vowel_diacritic.values
self.consonant_diacritic = df.consonant_diacritic
if len(folds) == 1:
self.aug = albumentations.Compose([
albumentations.Resize(img_height, img_width),
albumentations.Normalize(mean, std, always_apply=True)
])
else:
self.aug = albumentations.Compose([
albumentations.Resize(img_height, img_width),
albumentations.ShiftScaleRotate(shift_limit=0.0625,
scale_limit=0.1,
rotate_limit=5,
p=0.9),
albumentations.Rotate(limit=5),
albumentations.RandomContrast(limit=0.2),
albumentations.GaussianBlur(blur_limit=7),
albumentations.RandomGamma(),
albumentations.RandomShadow(),
albumentations.GaussNoise(),
albumentations.ChannelShuffle(),
#albumentations.Cutout(),
albumentations.Equalize(),
albumentations.MultiplicativeNoise(),
albumentations.Normalize(mean, std, always_apply=True)
])
def __init__(
self,
transforms=None,
mean=(0, 0, 0),
std=(1, 1, 1),
width=3,
depth=-1,
alpha=1.,
p=1.,
):
self.transforms = transforms
self.mean = mean
self.std = std
self.width = width
self.depth = depth
self.alpha = alpha
self.p = p
if self.transforms is None:
self.transforms = [
AutoContrast(cutoff=0, p=1),
albu.Equalize(mode='pil', p=1),
albu.Posterize(num_bits=(3, 4), p=1),
albu.ShiftScaleRotate(shift_limit=0, scale_limit=0, rotate_limit=5, border_mode=cv2.BORDER_CONSTANT, value=(0, 0, 0), p=1), # rotate
albu.Solarize(threshold=77, p=1),
RandomShear(shear_x=0.09, shear_y=0, border_mode=cv2.BORDER_CONSTANT, value=(0, 0, 0), p=1),
RandomShear(shear_x=0, shear_y=0.09, border_mode=cv2.BORDER_CONSTANT, value=(0, 0, 0), p=1),
VerticalShift(shift_limit=0.09, border_mode=cv2.BORDER_CONSTANT, value=(0, 0, 0), p=1),
HorizontalShift(shift_limit=0.09, border_mode=cv2.BORDER_CONSTANT, value=(0, 0, 0), p=1),
# ImageNet-C
albu.HueSaturationValue(hue_shift_limit=0, sat_shift_limit=(-36, 0), val_shift_limit=0, p=1), # saturation
albu.RandomContrast(limit=(-0.36, 0), p=1),
albu.RandomBrightness(limit=(-0.36, 0), p=1),
albu.OneOf([ # sharpness
albu.IAASharpen(alpha=(0.1, 0.5), lightness=0, p=1),
albu.Blur(blur_limit=7, p=1),
], p=0.5),
]
import albumentations as A
from albumentations.pytorch import ToTensorV2
train_transformation = A.Compose([
A.HorizontalFlip(p=0.5),
A.ShiftScaleRotate(rotate_limit=(-15, -15), border_mode=1, p=0.5),
A.OneOf([
A.RandomBrightnessContrast(p=0.5,
brightness_limit=(-0.1, 0.1),
contrast_limit=(-0.1, 0.1),
brightness_by_max=True),
A.Equalize(p=0.5, mode='cv', by_channels=True),
],
p=0.5),
A.RandomCrop(224, 224, p=1.0),
ToTensorV2(p=1.0)
])
test_transformation = A.Compose([A.Resize(224, 224, p=1.0), ToTensorV2(p=1.0)])
alb.OneOf([
alb.OpticalDistortion(p=0.3),
alb.GridDistortion(p=0.1),
alb.IAAAffine(p=0.1),
alb.ShiftScaleRotate(rotate_limit=15, p=0.1)
],
p=0.2),
alb.OneOf([alb.CLAHE(clip_limit=2),
alb.IAASharpen(),
alb.IAAEmboss()],
p=0.3),
alb.OneOf([
alb.HueSaturationValue(p=0.3),
alb.RGBShift(),
alb.RandomBrightnessContrast(),
alb.Equalize(mode='cv', p=1)
],
p=0.3),
alb.HorizontalFlip(),
alb.RandomGamma(p=0.2),
alb.OneOf([
alb.CoarseDropout(2, 10, 10, p=0.5),
alb.RandomRain(blur_value=3, p=1)
],
p=0.5)
]))
class MyDataset(Dataset):
def __init__(self,
data_dir,
def augment_and_save_images(self):
_, _, files = next(os.walk(self.img_dir_path))
counter = 1
rotate_and_flip = A.Compose(
[A.VerticalFlip(p=1), A.RandomRotate90(p=1)])
elastic_aug = A.ElasticTransform(p=1,
alpha=400,
sigma=20,
alpha_affine=0.5,
border_mode=cv2.BORDER_REPLICATE,
interpolation=cv2.INTER_LINEAR)
clahe_aug = A.CLAHE(clip_limit=(4.0, 16.0),
tile_grid_size=(4, 4),
always_apply=False,
p=1)
blur_aug = A.Blur(blur_limit=(5, 8), always_apply=False, p=1)
downscale_aug = A.Downscale(scale_min=0.25,
scale_max=0.4,
interpolation=0,
always_apply=False,
p=1)
equalize_aug = A.Equalize(mode='cv',
by_channels=True,
mask=None,
always_apply=False,
p=1)
gaussnoise_aug = A.GaussNoise(var_limit=(50, 100),
mean=0,
always_apply=False,
p=1)
saturate_aug = A.HueSaturationValue(hue_shift_limit=30,
sat_shift_limit=30,
val_shift_limit=30,
p=1)
embos_aug = A.IAAEmboss(alpha=(0.5, 0.8),
strength=(0.5, 0.8),
always_apply=False,
p=1)
sharpen_aug = A.IAASharpen(alpha=(0.5, 0.8),
lightness=(0.8, 1),
always_apply=False,
p=1)
solarize_aug = A.Solarize(threshold=(100, 200),
always_apply=False,
p=1)
aug_dict = {
'rotate': rotate_and_flip,
'elastic': elastic_aug,
'clahe': clahe_aug,
'blur': blur_aug,
'downscale': downscale_aug,
'equalize': equalize_aug,
'noise': gaussnoise_aug,
'saturate': saturate_aug,
'embos': embos_aug,
'sharpen': sharpen_aug,
'solarize': solarize_aug
}
required_aug_list = [
'rotate', 'elastic', 'noise', 'saturate', 'solarize'
]
for filename in files:
image_path = self.img_dir_path + '/' + filename
image = imageio.imread(image_path)
label_path = self.label_dir_path + '/' + filename
label = np.asarray(Image.open(label_path))
for aug_type in required_aug_list:
self.save_augmented_images(aug_name=aug_type,
augmented_items=aug_dict[aug_type](
image=image, mask=label),
image_name=counter)
#Following 3 if loops are separated to get extra randomness
if bool(random.getrandbits(1)):
self.save_augmented_images(aug_name='clahe',
augmented_items=aug_dict['clahe'](
image=image, mask=label),
image_name=counter)
else:
self.save_augmented_images(
aug_name='equalize',
augmented_items=aug_dict['equalize'](image=image,
mask=label),
image_name=counter)
if bool(random.getrandbits(1)):
self.save_augmented_images(aug_name='blur',
augmented_items=aug_dict['blur'](
image=image, mask=label),
image_name=counter)
else:
self.save_augmented_images(
aug_name='downscale',
augmented_items=aug_dict['downscale'](image=image,
mask=label),
image_name=counter)
if bool(random.getrandbits(1)):
self.save_augmented_images(aug_name='embos',
augmented_items=aug_dict['embos'](
image=image, mask=label),
image_name=counter)
else:
self.save_augmented_images(aug_name='sharpen',
augmented_items=aug_dict['sharpen'](
image=image, mask=label),
image_name=counter)
counter += 1
def get_transform_imagenet(use_albu_aug):
if use_albu_aug:
train_transform = al.Compose([
# al.Flip(p=0.5),
al.Resize(256, 256, interpolation=2),
al.RandomResizedCrop(224,
224,
scale=(0.08, 1.0),
ratio=(3. / 4., 4. / 3.),
interpolation=2),
al.HorizontalFlip(),
al.OneOf(
[
al.OneOf(
[
al.ShiftScaleRotate(
border_mode=cv2.BORDER_CONSTANT,
rotate_limit=30), # , p=0.05),
al.OpticalDistortion(
border_mode=cv2.BORDER_CONSTANT,
distort_limit=5.0,
shift_limit=0.1),
# , p=0.05),
al.GridDistortion(border_mode=cv2.BORDER_CONSTANT
), # , p=0.05),
al.ElasticTransform(
border_mode=cv2.BORDER_CONSTANT,
alpha_affine=15), # , p=0.05),
],
p=0.1),
al.OneOf(
[
al.RandomGamma(), # p=0.05),
al.HueSaturationValue(), # p=0.05),
al.RGBShift(), # p=0.05),
al.CLAHE(), # p=0.05),
al.ChannelShuffle(), # p=0.05),
al.InvertImg(), # p=0.05),
],
p=0.1),
al.OneOf(
[
al.RandomSnow(), # p=0.05),
al.RandomRain(), # p=0.05),
al.RandomFog(), # p=0.05),
al.RandomSunFlare(num_flare_circles_lower=1,
num_flare_circles_upper=2,
src_radius=110),
# p=0.05, ),
al.RandomShadow(), # p=0.05),
],
p=0.1),
al.RandomBrightnessContrast(p=0.1),
al.OneOf(
[
al.GaussNoise(), # p=0.05),
al.ISONoise(), # p=0.05),
al.MultiplicativeNoise(), # p=0.05),
],
p=0.1),
al.OneOf(
[
al.ToGray(), # p=0.05),
al.ToSepia(), # p=0.05),
al.Solarize(), # p=0.05),
al.Equalize(), # p=0.05),
al.Posterize(), # p=0.05),
al.FancyPCA(), # p=0.05),
],
p=0.1),
al.OneOf(
[
# al.MotionBlur(blur_limit=1),
al.Blur(blur_limit=[3, 5]),
al.MedianBlur(blur_limit=[3, 5]),
al.GaussianBlur(blur_limit=[3, 5]),
],
p=0.1),
al.OneOf(
[
al.CoarseDropout(), # p=0.05),
al.Cutout(), # p=0.05),
al.GridDropout(), # p=0.05),
al.ChannelDropout(), # p=0.05),
al.RandomGridShuffle(), # p=0.05),
],
p=0.1),
al.OneOf(
[
al.Downscale(), # p=0.1),
al.ImageCompression(quality_lower=60), # , p=0.1),
],
p=0.1),
],
p=0.5),
al.Normalize(),
ToTensorV2()
])
else:
train_transform = transforms.Compose([
transforms.Resize(256),
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
])
test_transform = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
])
if use_albu_aug:
train_transform = MultiDataTransformAlbu(train_transform)
else:
train_transform = MultiDataTransform(train_transform)
return train_transform, test_transform
aug_seq1 = A.OneOf([
A.Rotate(limit=(-90, 90), p=1.0),
A.Flip(p=1.0),
], p=1.0)
aug_seq2 = A.OneOf([
# A.ChannelDropout(always_apply=False, p=1.0, channel_drop_range=(1, 1), fill_value=0),
A.RGBShift(r_shift_limit=15, g_shift_limit=15, b_shift_limit=15, p=1.0),
A.RandomBrightnessContrast(always_apply=False, p=1.0, brightness_limit=(-0.2, 0.2), contrast_limit=(-0.2, 0.2), brightness_by_max=True)
], p=1.0)
aug_seq3 = A.OneOf([
A.GaussNoise(always_apply=False, p=1.0, var_limit=(10, 100)),
A.ISONoise(always_apply=False, p=1.0, intensity=(0.1, 1.0), color_shift=(0.01, 0.3)),
A.MultiplicativeNoise(always_apply=False, p=1.0, multiplier=(0.8, 1.6), per_channel=True, elementwise=True),
], p=1.0)
aug_seq4 = A.OneOf([
A.Equalize(always_apply=False, p=1.0, mode='pil', by_channels=True),
A.InvertImg(always_apply=False, p=1.0),
A.MotionBlur(always_apply=False, p=1.0, blur_limit=(3, 7)),
A.OpticalDistortion(always_apply=False, p=1.0, distort_limit=(-0.3, 0.3), shift_limit=(-0.05, 0.05), interpolation=0, border_mode=0, value=(0, 0, 0), mask_value=None),
A.RandomFog(always_apply=False, p=1.0, fog_coef_lower=0.1, fog_coef_upper=0.45, alpha_coef=0.5)
], p=1.0)
aug_seq = A.Compose([
A.Resize(IMG_SIZE, IMG_SIZE),
aug_seq1,
aug_seq2,
aug_seq3,
aug_seq4,
A.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])
aug_path = '/home/jitesh/prj/classification/test/bolt/aug/aug_seq.json'
A.save(aug_seq, aug_path)
def get_train_aug(RESOLUTION=300):
return A.Compose([
A.LongestMaxSize(max_size=RESOLUTION*2, interpolation=cv2.INTER_CUBIC, \
always_apply=True),
A.PadIfNeeded(min_height=RESOLUTION*2, min_width=RESOLUTION*2, always_apply=True, border_mode=cv2.BORDER_CONSTANT),
A.RandomResizedCrop(RESOLUTION,RESOLUTION, scale=(0.7, 1), \
interpolation=cv2.INTER_CUBIC),
A.Resize(RESOLUTION, RESOLUTION, p=1.0, interpolation=cv2.INTER_CUBIC),
A.FancyPCA(p=0.8, alpha=0.5),
# A.Transpose(p=0.7),
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.1),
A.ShiftScaleRotate(p=0.4, rotate_limit=12),
A.HueSaturationValue(
always_apply=False, p=0.3,
hue_shift_limit=(-20, 20),
sat_shift_limit=(-30, 30),
val_shift_limit=(-20, 20)),
# A.HueSaturationValue(
# hue_shift_limit=0.4, #.3
# sat_shift_limit=0.4, #.3
# val_shift_limit=0.4, #.3
# p=0.7
# ),
A.RandomBrightnessContrast(
brightness_limit=(-0.5,0.5), #-.2,.2
contrast_limit=(-0.4, 0.4), #-.2,.2
#p=0.6
),
A.CoarseDropout(p=0.8, max_holes=30),
# A.Cutout(p=0.8, max_h_size=40, max_w_size=40),
A.Cutout(p=1, max_h_size=60, max_w_size=30, num_holes=6, fill_value=[106,87,55]),
A.Cutout(p=1, max_h_size=30, max_w_size=60, num_holes=6, fill_value=[106,87,55]),
A.OneOf([
A.OpticalDistortion(always_apply=False, p=1.0, distort_limit=(-0.6599999666213989, 0.6800000071525574),
shift_limit=(-0.6699999570846558, 0.4599999785423279), interpolation=0,
border_mode=0, value=(0, 0, 0), mask_value=None),
# A.OpticalDistortion(p=0.5, distort_limit=0.15, shift_limit=0.15),
# A.GridDistortion(p=0.5, distort_limit=0.5),
A.GridDistortion(always_apply=False, p=1.0,
num_steps=6, distort_limit=(-0.4599999785423279, 0.5),
interpolation=0, border_mode=0,
value=(0, 0, 0), mask_value=None),
# A.IAAPiecewiseAffine(p=0.5, scale=(0.1, 0.14)),
], p=0.6),
A.Sharpen(p=1.0, alpha=(0.1,0.3), lightness=(0.3, 0.9)),
A.GaussNoise(var_limit=(300.0, 500.0), p=0.4),
A.ISONoise(always_apply=False, p=0.4,
intensity=(0.10000000149011612, 1.399999976158142),
color_shift=(0.009999999776482582, 0.4000000059604645)),
A.OneOf([
A.Equalize(always_apply=False, p=1.0, mode='cv', by_channels=True),
A.Solarize(always_apply=False, p=1.0, threshold=(67, 120)),
# A.IAAAdditiveGaussianNoise(p=1.0),
A.GaussNoise(p=1.0),
A.MotionBlur(always_apply=False, p=1.0, blur_limit=(5, 20))
], p=0.5),
], p=1.0)
rotate_limit=18, p=1, border_mode=cv.BORDER_CONSTANT),
A.IAAAffine(shear=10, p=1, mode="constant"),
#A.Perspective(scale=(0.05, 0.15), keep_size=True, pad_mode=0, pad_val=0,
# mask_pad_val=0, fit_output=False, interpolation=1, always_apply=False, p=1),
],
p=1.0,
),
A.OneOf(
[
A.FancyPCA(alpha=0.1, always_apply=False, p=1),
A.Blur(p=1),
A.ToGray(p=0.8),
A.ColorJitter(
brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1, p=1),
A.ChannelDropout((1, 1), fill_value=0, always_apply=False, p=1),
],
p=0.3,
),
A.OneOf(
[ #A.GaussNoise (var_limit=(10.0, 50.0), mean=0, per_channel=True, always_apply=False, p=0.5),
A.Equalize(mode='cv',
by_channels=True,
mask=None,
mask_params=(),
always_apply=False,
p=0.8),
A.MotionBlur(blur_limit=4, p=1),
],
p=0.1,
)
])
def __call__(self, image, boxes=None, labels=None):
#initialize the format for lib albumentations
if boxes.shape[0] == 0:
return image, boxes, labels
bbox = []
for i in boxes:
bbox.append(list(i))
#create annotations
annotations = {
'image': image,
'bboxes': boxes,
'category_id': list(labels)
}
#create translation
#Color_Level Change
if self.cfg.DATA_LOADER.AUGMENTATION_WEATHER:
trans_color_level = A.Compose([
A.Cutout(num_holes=20,
max_h_size=64,
max_w_size=64,
fill_value=255,
always_apply=False,
p=0.8),
A.Equalize(p=1),
A.HueSaturationValue(hue_shift_limit=50,
sat_shift_limit=50,
val_shift_limit=50,
always_apply=False,
p=0.8),
A.OneOf([
A.RandomFog(fog_coef_lower=0.3,
fog_coef_upper=0.7,
alpha_coef=0.08,
always_apply=False,
p=0.5),
A.RandomSnow(snow_point_lower=0.1,
snow_point_upper=0.3,
brightness_coeff=2.5,
always_apply=False,
p=0.5),
A.RandomSunFlare(flare_roi=(0, 0, 1, 0.5),
angle_lower=0,
angle_upper=1,
num_flare_circles_lower=6,
num_flare_circles_upper=10,
src_radius=400,
src_color=(255, 255, 255),
always_apply=False,
p=0.5),
A.RandomRain(slant_lower=-10,
slant_upper=10,
drop_length=20,
drop_width=1,
drop_color=(200, 200, 200),
blur_value=7,
brightness_coefficient=0.7,
rain_type=None,
always_apply=False,
p=0.5)
]),
A.OneOf([
A.RandomSizedBBoxSafeCrop(720,
960,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
A.RandomSizedBBoxSafeCrop(480,
640,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
A.RandomSizedBBoxSafeCrop(240,
320,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
]),
])
else:
trans_color_level = A.Compose([
A.Cutout(num_holes=20,
max_h_size=64,
max_w_size=64,
fill_value=255,
always_apply=False,
p=0.5),
A.Equalize(p=1),
A.HueSaturationValue(hue_shift_limit=50,
sat_shift_limit=50,
val_shift_limit=50,
always_apply=False,
p=0.5),
A.OneOf([
A.RandomSizedBBoxSafeCrop(720,
960,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
A.RandomSizedBBoxSafeCrop(480,
640,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
A.RandomSizedBBoxSafeCrop(240,
320,
erosion_rate=0.0,
interpolation=1,
always_apply=False,
p=0.5),
]),
])
#Spatial_Level
if self.cfg.DATA_LOADER.AUGMENTATION_SPATIAL_LEVEL:
trans_rotate_level = A.Compose([
A.OneOf([
A.Rotate(limit=90,
interpolation=1,
border_mode=4,
value=None,
mask_value=None,
always_apply=False,
p=0.5),
A.RandomRotate90(always_apply=False, p=0.5),
A.VerticalFlip(always_apply=False, p=0.5),
A.HorizontalFlip(always_apply=False, p=0.5)
]),
])
#Apply the trans
aug = get_aug(trans_color_level)
augmented = aug(**annotations)
img = augmented['image']
bbox = augmented['bboxes']
bbox = np.array(bbox)
label = augmented['category_id']
#try rotate
if self.cfg.DATA_LOADER.AUGMENTATION_SPATIAL_LEVEL:
aug1 = get_aug(trans_rotate_level)
augmented1 = aug1(**augmented)
img1 = augmented1['image']
bbox1 = augmented1['bboxes']
bbox1 = np.array(bbox1)
label1 = augmented1['category_id']
#if rotate fail
if bbox1.shape[0] == 0:
return img, bbox.astype(np.float32), np.array(label)
else:
return img1, bbox1.astype(np.float32), np.array(label1)
else:
return img, bbox.astype(np.float32), np.array(label)
def set_transforms(self) -> dict:
d4_tansforms = [
A.SmallestMaxSize(self.img_size[0], interpolation=0, p=1.),
# D4 Group augmentations
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
A.Transpose(p=0.5),
]
tensor_norm = [
ToTensor(),
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
]
geometric = [
A.SmallestMaxSize(self.img_size[0], interpolation=0, p=1.),
A.ShiftScaleRotate(shift_limit=self.shift_limit,
self.scale_limit,
rotate_limit=self.rotate_limit,
interpolation=cv2.INTER_LINEAR,
border_mode=cv2.BORDER_CONSTANT,
value=0,
mask_value=0,
p=0.5),
# crop and resize
A.RandomSizedCrop(
(self.crop_size[0],
min(self.crop_size[1], self.img_size[0], self.img_size[1])),
self.img_size[0],
self.img_size[1],
w2h_ratio=1.0,
interpolation=cv2.INTER_LINEAR,
always_apply=False,
p=0.2),
]
resize_crop = [
A.SmallestMaxSize(max(self.img_size[0], self.img_size[1]),
interpolation=0,
p=1.),
A.RandomCrop(self.img_size[0], self.img_size[1], p=1.),
]
train_light = [
A.SmallestMaxSize(self.img_size[0], interpolation=0, p=1.),
A.RandomCrop(self.img_size[0], self.img_size[1], p=1.),
A.HorizontalFlip(p=0.5),
A.RandomBrightnessContrast(p=0.5),
]
train_medium = train_light.append([
A.ShiftScaleRotate(shift_limit=self.shift_limit,
self.scale_limit,
rotate_limit=self.rotate_limit,
p=0.5),
A.OneOf([
A.RandomBrightnessContrast(p=0.7),
A.Equalize(p=0.3),
A.HueSaturationValue(p=0.5),
A.RGBShift(p=0.5),
A.RandomGamma(p=0.4),
A.ChannelShuffle(p=0.05),
],
p=0.9),
A.OneOf([
A.GaussNoise(p=0.5),
A.ISONoise(p=0.5),
A.MultiplicativeNoise(0.5),
],
p=0.2),
])
valid_ade = [
A.SmallestMaxSize(self.img_size, p=1.),
A.Lambda(name="Pad32", image=pad_x32, mask=pad_x32),
]
# from bloodaxe
# https://github.com/BloodAxe/Catalyst-Inria-Segmentation-Example/blob/master/inria/augmentations.py
crop = [ #(image_size: Tuple[int, int], min_scale=0.75, max_scale=1.25, input_size=5000):
A.OneOrOther(
A.RandomSizedCrop((self.crop_size[0],
min(self.crop_size[1], self.img_size[0],
self.img_size[1])), self.img_size[0],
self.img_size[1]),
A.CropNonEmptyMaskIfExists(self.img_size[0], self.img_size[1]),
)
]
safe_augmentations = [A.HorizontalFlip(), A.RandomBrightnessContrast()]
light_augmentations = [
A.HorizontalFlip(),
A.RandomBrightnessContrast(),
A.OneOf([
A.ShiftScaleRotate(scale_limit=self.scale_limit,
rotate_limit=self.rotate_limit,
border_mode=cv2.BORDER_CONSTANT),
A.IAAAffine(),
A.IAAPerspective(),
A.NoOp()
]),
A.HueSaturationValue(),
]
medium_augmentations = [
A.HorizontalFlip(),
A.ShiftScaleRotate(scale_limit=self.scale_limit,
rotate_limit=self.rotate_limit,
border_mode=cv2.BORDER_CONSTANT),
# Add occasion blur/sharpening
A.OneOf([A.GaussianBlur(),
A.IAASharpen(),
A.NoOp()]),
# Spatial-preserving augmentations:
A.OneOf(
[A.CoarseDropout(),
A.MaskDropout(max_objects=5),
A.NoOp()]),
A.GaussNoise(),
A.OneOf([
A.RandomBrightnessContrast(),
A.CLAHE(),
A.HueSaturationValue(),
A.RGBShift(),
A.RandomGamma()
]),
# Weather effects
A.RandomFog(fog_coef_lower=0.01, fog_coef_upper=0.3, p=0.1),
]
hard_augmentations = [
A.RandomRotate90(),
A.Transpose(),
A.RandomGridShuffle(),
A.ShiftScaleRotate(scale_limit=self.scale_limit,
rotate_limit=self.rotate_limit,
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.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()
]),
]
TRANSFORMS = {
"d4": d4_tansforms,
"resize_crop": resize_crop,
"geometric": geometric,
"light": train_light,
"medium": train_medium,
"ade_valid": valid_ade,
"flip_bright": safe_augmentations,
"crop": crop,
"inria_light": light_augmentations,
"inria_medium": medium_augmentations,
"inria_hard": hard_augmentations,
"inria_valid": safe_augmentations,
}
return TRANSFORMS
def get_transforms(self):
augs = []
if self.augs_name:
augs = self.set_transforms[self.augs_name]
augs.append(
A.Resize(height=self.img_size[0],
width=self.img_size[1],
p=1.0))
if self.use_d4:
augs.append(self.set_transforms["d4"])
if self.vflip:
augs.append(A.VerticalFlip(p=0.5))
if self.hflip:
augs.append(A.HorizontalFlip(p=0.5))
if self.normalise:
augs.append(A.Normalize())
return A.Compose(augs)
def __init__(self,
path,
img_size=640,
batch_size=16,
augment=False,
hyp=None,
rect=False,
image_weights=False,
cache_images=False,
single_cls=False,
stride=32,
pad=0.0,
rank=-1):
self.img_size = img_size
self.augment = augment
self.hyp = hyp
self.image_weights = image_weights
self.rect = False if image_weights else rect
self.mosaic = self.augment and not self.rect # load 4 images at a time into a mosaic (only during training)
self.mosaic_border = [-img_size // 2, -img_size // 2]
self.stride = stride
self.albuAug = A.Compose([
A.Blur((3, 6), p=0.07),
A.CLAHE((1, 4), (8, 8), p=0.07),
A.ChannelShuffle(p=0),
A.Equalize(always_apply=False, p=0.07, mode='cv',
by_channels=True),
A.GaussNoise(always_apply=False,
p=0.07,
var_limit=(144.739990234375, 266.4499816894531)),
A.HueSaturationValue(always_apply=False,
p=0.07,
hue_shift_limit=(-20, 20),
sat_shift_limit=(-30, 30),
val_shift_limit=(-20, 20)),
A.ISONoise(always_apply=False,
p=0.07,
intensity=(0.1, 0.5),
color_shift=(0.01, 0.05)),
A.ImageCompression(always_apply=False,
p=0.07,
quality_lower=25,
quality_upper=66,
compression_type=1),
A.JpegCompression(always_apply=False,
p=0.07,
quality_lower=23,
quality_upper=45),
A.MotionBlur(always_apply=False, p=0.07, blur_limit=(3, 7)),
A.MultiplicativeNoise(always_apply=False,
p=0.07,
multiplier=(0.9, 1.1),
per_channel=True,
elementwise=True),
A.RandomBrightness(always_apply=False, p=0.07, limit=(-0.2, 0.2)),
A.RandomBrightnessContrast(always_apply=False,
p=0.07,
brightness_limit=(-0.2, 0.2),
contrast_limit=(-0.2, 0.2),
brightness_by_max=True)
])
try:
f = [] # image files
for p in path if isinstance(path, list) else [path]:
p = Path(p) # os-agnostic
if p.is_dir(): # dir
f += glob.glob(str(p / '**' / '*.*'), recursive=True)
elif p.is_file(): # file
with open(p, 'r') as t:
t = t.read().strip().splitlines()
parent = str(p.parent) + os.sep
f += [
x.replace('./', parent)
if x.startswith('./') else x for x in t
] # local to global path
else:
raise Exception('%s does not exist' % p)
self.img_files = sorted([
x.replace('/', os.sep) for x in f
if x.split('.')[-1].lower() in img_formats
])
assert self.img_files, 'No images found'
except Exception as e:
raise Exception('Error loading data from %s: %s\nSee %s' %
(path, e, help_url))
# Check cache
self.label_files = img2label_paths(self.img_files) # labels
cache_path = Path(self.label_files[0]).parent.with_suffix(
'.cache') # cached labels
if cache_path.is_file():
cache = torch.load(cache_path) # load
if cache['hash'] != get_hash(
self.label_files +
self.img_files) or 'results' not in cache: # changed
cache = self.cache_labels(cache_path) # re-cache
else:
cache = self.cache_labels(cache_path) # cache
# Display cache
[nf, nm, ne, nc,
n] = cache.pop('results') # found, missing, empty, corrupted, total
desc = f"Scanning '{cache_path}' for images and labels... {nf} found, {nm} missing, {ne} empty, {nc} corrupted"
tqdm(None, desc=desc, total=n, initial=n)
assert nf > 0 or not augment, f'No labels found in {cache_path}. Can not train without labels. See {help_url}'
# Read cache
cache.pop('hash') # remove hash
labels, shapes = zip(*cache.values())
self.labels = list(labels)
self.shapes = np.array(shapes, dtype=np.float64)
self.img_files = list(cache.keys()) # update
self.label_files = img2label_paths(cache.keys()) # update
if single_cls:
for x in self.labels:
x[:, 0] = 0
n = len(shapes) # number of images
bi = np.floor(np.arange(n) / batch_size).astype(np.int) # batch index
nb = bi[-1] + 1 # number of batches
self.batch = bi # batch index of image
self.n = n
self.indices = range(n)
# Rectangular Training
if self.rect:
# Sort by aspect ratio
s = self.shapes # wh
ar = s[:, 1] / s[:, 0] # aspect ratio
irect = ar.argsort()
self.img_files = [self.img_files[i] for i in irect]
self.label_files = [self.label_files[i] for i in irect]
self.labels = [self.labels[i] for i in irect]
self.shapes = s[irect] # wh
ar = ar[irect]
# Set training image shapes
shapes = [[1, 1]] * nb
for i in range(nb):
ari = ar[bi == i]
mini, maxi = ari.min(), ari.max()
if maxi < 1:
shapes[i] = [maxi, 1]
elif mini > 1:
shapes[i] = [1, 1 / mini]
self.batch_shapes = np.ceil(
np.array(shapes) * img_size / stride + pad).astype(
np.int) * stride
# Cache images into memory for faster training (WARNING: large datasets may exceed system RAM)
self.imgs = [None] * n
if cache_images:
gb = 0 # Gigabytes of cached images
self.img_hw0, self.img_hw = [None] * n, [None] * n
results = ThreadPool(8).imap(lambda x: load_image(*x),
zip(repeat(self),
range(n))) # 8 threads
pbar = tqdm(enumerate(results), total=n)
for i, x in pbar:
self.imgs[i], self.img_hw0[i], self.img_hw[
i] = x # img, hw_original, hw_resized = load_image(self, i)
gb += self.imgs[i].nbytes
pbar.desc = 'Caching images (%.1fGB)' % (gb / 1E9)
def get_augmentation(version):
if version == "v1":
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v2":
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.OneOf([
A.RandomCrop(height=256, width=256, p=1),
A.RandomCrop(height=384, width=384, p=1),
A.RandomCrop(height=512, width=512, p=1),
A.RandomCrop(height=640, width=640, p=1),
],
p=0.5),
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v3":
return {
"train":
A.Compose([
A.RandomCrop(height=1024, width=1024, p=1),
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v4":
# for 3x-scaled dataset | 256x256
return {
"train":
A.Compose([
A.RandomCrop(height=512, width=512, p=1),
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v5":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.RandomCrop(height=512, width=512, p=1),
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v6":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.25),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.25),
A.RandomContrast(limit=0.2, always_apply=False, p=0.25),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.25),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v7":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.RandomCrop(height=768, width=768, p=1),
A.Resize(height=256, width=256, p=1.0),
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.25),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.25),
A.RandomContrast(limit=0.2, always_apply=False, p=0.25),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.25),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Resize(height=256, width=256, p=1.0),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v8":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Rotate(p=0.5),
A.RandomCrop(height=512, width=512, p=1),
A.Flip(p=0.5),
# color
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.25),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.25),
A.RandomContrast(limit=0.2, always_apply=False, p=0.25),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.25),
# transform
A.ElasticTransform(p=1, alpha=120, sigma=6, alpha_affine=0.25),
A.GridDistortion(p=0.25),
A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=0.25),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v9":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.25),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.25),
A.RandomContrast(limit=0.2, always_apply=False, p=0.25),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.25),
# transform
A.OneOf([
A.ElasticTransform(
alpha=120, sigma=6, alpha_affine=0.25, p=0.25),
A.GridDistortion(p=0.25),
A.OpticalDistortion(
distort_limit=2, shift_limit=0.5, p=0.25)
],
p=0.75),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v10":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
# size
A.OneOf([
A.CropNonEmptyMaskIfExists(height=128, width=128, p=1.0),
A.CropNonEmptyMaskIfExists(height=256, width=256, p=1.0),
A.CropNonEmptyMaskIfExists(height=384, width=384, p=1.0)
],
p=0.25),
A.PadIfNeeded(min_height=512,
min_width=512,
border_mode=4,
p=1),
# array shuffle
A.OneOf([A.MaskDropout(p=1),
A.RandomGridShuffle(p=1)], p=0.5),
# quality
A.Downscale(scale_min=0.25,
scale_max=0.75,
interpolation=0,
always_apply=False,
p=0.5),
A.GaussNoise(p=0.5),
A.OneOf([
A.GlassBlur(p=1),
A.GaussianBlur(p=1),
], p=0.5),
# colors
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.5),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.5),
A.RandomContrast(limit=0.2, always_apply=False, p=0.5),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.5),
A.CLAHE(p=0.5),
A.Equalize(p=0.5),
A.ChannelShuffle(p=0.5),
# transform
A.OneOf([
A.ElasticTransform(
alpha=120, sigma=6, alpha_affine=0.25, p=1),
A.GridDistortion(p=1),
A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=1)
],
p=0.5),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v11":
# custom normalization (see other/img_normalization.py)
return {
"train":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
elif version == "v12":
# for 3x-scaled dataset | 512x512
return {
"train":
A.Compose([
A.Flip(p=0.5),
A.Rotate(p=0.5),
# size / quality
A.OneOf([
A.CropNonEmptyMaskIfExists(height=128, width=128, p=1.0),
A.CropNonEmptyMaskIfExists(height=256, width=256, p=1.0),
A.CropNonEmptyMaskIfExists(height=384, width=384, p=1.0),
A.Downscale(scale_min=0.25,
scale_max=0.75,
interpolation=0,
always_apply=False,
p=1.0),
],
p=0.25),
A.PadIfNeeded(min_height=512,
min_width=512,
border_mode=4,
p=1),
# array shuffle
A.OneOf([A.MaskDropout(p=1),
A.RandomGridShuffle(p=1)], p=0.15),
# noise
A.OneOf([
A.GaussNoise(p=1),
A.GlassBlur(p=1),
A.GaussianBlur(p=1),
],
p=0.15),
# colors
A.HueSaturationValue(hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=20,
always_apply=False,
p=0.15),
A.RandomBrightness(limit=0.4, always_apply=False, p=0.15),
A.RandomContrast(limit=0.2, always_apply=False, p=0.15),
A.RandomShadow(shadow_roi=(0, 0, 1, 1),
num_shadows_lower=1,
num_shadows_upper=50,
shadow_dimension=5,
always_apply=False,
p=0.15),
A.OneOf([
A.CLAHE(p=1),
A.Equalize(p=1),
A.ChannelShuffle(p=1),
],
p=0.15),
# transform
A.OneOf([
A.ElasticTransform(
alpha=120, sigma=6, alpha_affine=0.25, p=1),
A.GridDistortion(p=1),
A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=1)
],
p=0.15),
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
"valid":
A.Compose([
A.Normalize(mean=(0.623, 0.520, 0.650),
std=(0.278, 0.305, 0.274),
max_pixel_value=255.0,
p=1.0)
]),
}
else:
raise Exception(f"Augmentation version '{version}' is UNKNOWN!")
def get_train_transforms_mmdetection(input_size,
use_crop=False,
use_no_color_aug=False,
use_center_crop=False,
center_crop_ratio=0.9,
use_gray=False):
if isinstance(input_size, int):
input_size = (input_size[0], input_size[1])
return al.Compose([
al.RandomResizedCrop(height=input_size[0],
width=input_size[1],
scale=(0.4, 1.0),
interpolation=0,
p=0.5),
al.Resize(input_size[0], input_size[1], p=1.0),
al.HorizontalFlip(p=0.5),
al.OneOf([
al.ShiftScaleRotate(border_mode=0,
shift_limit=(-0.2, 0.2),
scale_limit=(-0.2, 0.2),
rotate_limit=(-20, 20)),
al.OpticalDistortion(border_mode=0,
distort_limit=[-0.5, 0.5],
shift_limit=[-0.5, 0.5]),
al.GridDistortion(
num_steps=5, distort_limit=[-0., 0.3], border_mode=0),
al.ElasticTransform(border_mode=0),
al.IAAPerspective(),
al.RandomGridShuffle()
],
p=0.1),
al.Rotate(limit=(-25, 25), border_mode=0, p=0.1),
al.OneOf([
al.RandomBrightnessContrast(brightness_limit=(-0.2, 0.2),
contrast_limit=(-0.2, 0.2)),
al.HueSaturationValue(hue_shift_limit=(-20, 20),
sat_shift_limit=(-30, 30),
val_shift_limit=(-20, 20)),
al.RandomGamma(gamma_limit=(30, 150)),
al.RGBShift(),
al.CLAHE(clip_limit=(1, 15)),
al.ChannelShuffle(),
al.InvertImg(),
],
p=0.1),
al.RandomSnow(p=0.05),
al.RandomRain(p=0.05),
al.RandomFog(p=0.05),
al.RandomSunFlare(num_flare_circles_lower=1,
num_flare_circles_upper=2,
src_radius=110,
p=0.05),
al.RandomShadow(p=0.05),
al.GaussNoise(var_limit=(10, 20), p=0.05),
al.ISONoise(color_shift=(0, 15), p=0.05),
al.MultiplicativeNoise(p=0.05),
al.OneOf([
al.ToGray(p=1. if use_gray else 0.05),
al.ToSepia(p=0.05),
al.Solarize(p=0.05),
al.Equalize(p=0.05),
al.Posterize(p=0.05),
al.FancyPCA(p=0.05),
],
p=0.05),
al.OneOf([
al.MotionBlur(blur_limit=(3, 7)),
al.Blur(blur_limit=(3, 7)),
al.MedianBlur(blur_limit=3),
al.GaussianBlur(blur_limit=3),
],
p=0.05),
al.CoarseDropout(p=0.05),
al.Cutout(num_holes=30,
max_h_size=37,
max_w_size=37,
fill_value=0,
p=0.05),
al.GridDropout(p=0.05),
al.ChannelDropout(p=0.05),
al.Downscale(scale_min=0.5, scale_max=0.9, p=0.1),
al.ImageCompression(quality_lower=60, p=0.2),
al.Normalize(),
ToTensorV2()
])
def equalize(m):
return albu.Equalize(
mode='pil',
p=1)
image = self.transforms(image=image)['image']
return image, torch.LongTensor([label])
df = pd.read_excel('SFEW.xlsx')
df = df.sample(frac=1.0, random_state=random_seed)
transforms_train = A.Compose([
A.HorizontalFlip(p=0.5),
A.Rotate(limit=15, p=0.5),
A.RandomResizedCrop(height=img_size,
width=img_size,
scale=(0.9, 1.0),
p=1.0),
A.Equalize(p=0.5),
A.Normalize(p=1.0),
ToTensorV2(p=1.0),
])
transforms_eval = A.Compose([
A.Resize(height=img_size, width=img_size, p=1.0),
A.Normalize(p=1.0),
ToTensorV2(p=1.0),
])
train_index = int(len(df) * train_portion)
valid_index = train_index + int(len(df) * valid_portion)
train = df.iloc[:train_index]
valid = df.iloc[train_index:valid_index]
test = df.iloc[valid_index:]
def get_train_transforms_atopy(input_size,
use_crop=False,
use_no_color_aug=False):
if use_crop:
resize = [
al.Resize(int(input_size * 1.2), int(input_size * 1.2)),
al.RandomSizedCrop(min_max_height=(int(input_size * 0.6),
int(input_size * 1.2)),
height=input_size,
width=input_size)
]
else:
resize = [al.Resize(input_size, input_size)]
return al.Compose(resize + [
al.Flip(p=0.5),
al.OneOf([
al.RandomRotate90(),
al.Rotate(limit=180),
], p=0.5),
al.OneOf([
al.ShiftScaleRotate(),
al.OpticalDistortion(),
al.GridDistortion(),
al.ElasticTransform(),
],
p=0.3),
al.RandomGridShuffle(p=0.05),
al.OneOf([
al.RandomGamma(),
al.HueSaturationValue(),
al.RGBShift(),
al.CLAHE(),
al.ChannelShuffle(),
al.InvertImg(),
],
p=0.1),
al.RandomSnow(p=0.05),
al.RandomRain(p=0.05),
al.RandomFog(p=0.05),
al.RandomSunFlare(p=0.05),
al.RandomShadow(p=0.05),
al.RandomBrightnessContrast(p=0.05),
al.GaussNoise(p=0.2),
al.ISONoise(p=0.2),
al.MultiplicativeNoise(p=0.2),
al.ToGray(p=0.05),
al.ToSepia(p=0.05),
al.Solarize(p=0.05),
al.Equalize(p=0.05),
al.Posterize(p=0.05),
al.FancyPCA(p=0.05),
al.OneOf([
al.MotionBlur(blur_limit=3),
al.Blur(blur_limit=3),
al.MedianBlur(blur_limit=3),
al.GaussianBlur(blur_limit=3),
],
p=0.05),
al.CoarseDropout(p=0.05),
al.Cutout(p=0.05),
al.GridDropout(p=0.05),
al.ChannelDropout(p=0.05),
al.Downscale(p=0.1),
al.ImageCompression(quality_lower=60, p=0.2),
al.Normalize(),
ToTensorV2()
])