def get_training_augmentation():
train_transform = [
A.RandomSizedCrop(min_max_height=(300, 360), height=320, width=320,
always_apply=True),
A.HorizontalFlip(p=0.5),
A.OneOf([
A.CLAHE(),
A.RandomBrightnessContrast(),
A.RandomGamma(),
A.HueSaturationValue(),
A.NoOp()
]),
A.OneOf([
A.IAAAdditiveGaussianNoise(p=0.2),
A.IAASharpen(),
A.Blur(blur_limit=3),
A.MotionBlur(blur_limit=3),
A.NoOp()
]),
A.OneOf([
A.RandomFog(),
A.RandomSunFlare(),
A.RandomRain(),
A.RandomSnow(),
A.NoOp()
]),
A.Normalize(),
]
return A.Compose(train_transform)
def medium_augmentations():
return A.Compose([
A.HorizontalFlip(),
A.ShiftScaleRotate(scale_limit=0.1,
rotate_limit=15,
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),
A.Normalize(),
])
def medium_3(image_size, p=1.0):
cutout_crop = int(0.25 * image_size)
return A.Compose(
[
# RandomCrop(input_size) / RandomResizedCrop (0.08, 1)
A.HorizontalFlip(p=0.5), # vflip
A.VerticalFlip(p=0.5), # hflip
A.ShiftScaleRotate(
shift_limit=0.0625, scale_limit=0.2, rotate_limit=45, p=0.3),
A.OneOf([
A.RandomFog(
fog_coef_lower=0.3, fog_coef_upper=1.0, alpha_coef=.1),
A.ImageCompression(quality_lower=20, quality_upper=99),
],
p=0.3),
A.RandomBrightnessContrast(brightness_limit=0.125,
contrast_limit=0.2,
p=0.5), # contrast_limit=0.5
A.HueSaturationValue(hue_shift_limit=5,
sat_shift_limit=30,
val_shift_limit=20,
p=0.2),
A.GaussNoise(var_limit=(1, 50), p=0.4),
A.CoarseDropout(min_holes=1,
max_holes=2,
max_height=cutout_crop,
max_width=cutout_crop,
p=0.5),
],
p=p)
def augment(self, img0, img1):
transform = A.Compose([
A.IAAAdditiveGaussianNoise(p=0.05),
A.OneOf([
A.IAASharpen(p=1.0),
A.Blur(blur_limit=3, p=1.0),
],
p=0.5),
A.OneOf([
A.RandomBrightnessContrast(p=1.0),
A.HueSaturationValue(p=1.0),
A.RandomGamma(p=1.0),
],
p=0.5),
A.OneOf([
A.RandomFog(p=1.0),
A.RandomRain(p=1.0),
A.RandomShadow(p=1.0),
A.RandomSnow(p=1.0),
A.RandomSunFlare(p=1.0)
],
p=0.05),
],
additional_targets={'img1': 'image'})
transformed = transform(image=img0, img1=img1)
img0 = transformed["image"]
img1 = transformed["img1"]
return img0, img1
def get_transformations(aug_name='soft', image_size=448):
all_transforms = {
'no_aug' : A.Compose([
A.Resize(image_size, image_size),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
]),
'super_soft' : A.Compose([
A.Resize(image_size, image_size),
A.HorizontalFlip(p=0.5),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
]),
'soft_rtsd' : A.Compose([
A.Resize(image_size, image_size),
A.RandomBrightnessContrast(brightness_limit=(-0.3,0.3), contrast_limit=(-0.3,0.3),p=0.5),
A.Cutout(num_holes=5, max_h_size=image_size//10, max_w_size=image_size//10, p=0.25),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
]),
'medium_rtsd' : A.Compose([
A.Resize(image_size, image_size),
A.RandomBrightnessContrast(brightness_limit=(-0.3,0.3), contrast_limit=(-0.3,0.3),p=0.5),
A.Blur(p=0.25, blur_limit=(3, 5)),
A.GaussNoise(p=0.25, var_limit=(10.0, 50.0)),
A.RGBShift(p=0.25, r_shift_limit=(-20, 20), g_shift_limit=(-20, 20), b_shift_limit=(-20, 20)),
A.RandomFog(p=0.1, fog_coef_lower=0.1, fog_coef_upper=0.44, alpha_coef=0.16),
A.Cutout(num_holes=5, max_h_size=image_size//10, max_w_size=image_size//10, p=0.25),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
]),
'soft' : A.Compose([
A.Resize(image_size, image_size),
A.HorizontalFlip(p=0.5),
A.ShiftScaleRotate(shift_limit=0.1, scale_limit=1.1, rotate_limit=15, p=0.5),
A.RandomBrightnessContrast(brightness_limit=(-0.2,0.2), contrast_limit=(-0.2,0.2),p=0.5),
A.Cutout(num_holes=8, max_h_size=image_size//5, max_w_size=image_size//5, p=0.5),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
]),
'transforms_without_aug' : A.Compose([
A.Resize(image_size, image_size),
A.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
ToTensorV2()
])
}
image_transforms = {
'train': all_transforms[aug_name],
'valid': all_transforms['transforms_without_aug'],
'test': all_transforms['transforms_without_aug']
}
return image_transforms
def generate_transforms2(img_size):
train_transform = Compose([
# A.RandomCrop(p=1, height=img_size, width=img_size),
A.Resize(height=img_size, width=img_size),
A.RandomSunFlare(p=1),
A.RandomFog(p=1),
A.RandomBrightness(p=1),
A.Rotate(p=1, limit=90),
A.RGBShift(p=1),
A.RandomSnow(p=1),
A.HorizontalFlip(p=1),
A.VerticalFlip(p=1),
A.RandomContrast(limit=0.5, p=1),
A.HueSaturationValue(p=1,
hue_shift_limit=20,
sat_shift_limit=30,
val_shift_limit=50),
# A.Cutout(p=1),
# A.Transpose(p=1),
A.JpegCompression(p=1),
A.CoarseDropout(p=1),
A.IAAAdditiveGaussianNoise(loc=0,
scale=(2.5500000000000003, 12.75),
per_channel=False,
p=1),
A.IAAAffine(scale=1.0,
translate_percent=None,
translate_px=None,
rotate=0.0,
shear=0.0,
order=1,
cval=0,
mode='reflect',
p=1),
A.IAAAffine(rotate=90., p=1),
A.IAAAffine(rotate=180., p=1),
Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225),
max_pixel_value=255.0,
p=1.0),
ToTensorV2()
])
val_transform = Compose([
Resize(height=img_size, width=img_size),
Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225),
max_pixel_value=255.0,
p=1.0),
ToTensorV2(),
])
return {"train": train_transform, "val": val_transform}
def get_training_augmentation(min_area=0., min_visibility=0.):
train_transform = [
albu.OneOf([
albu.MotionBlur(p=.2),
albu.MedianBlur(blur_limit=3, p=0.1),
albu.Blur(blur_limit=3, p=0.1),
],
p=0.2),
albu.ShiftScaleRotate(shift_limit=0,
scale_limit=0,
rotate_limit=15,
p=0.5),
albu.OneOf([
albu.CLAHE(clip_limit=2),
albu.IAASharpen(),
albu.IAAEmboss(),
],
p=0.3),
albu.OneOf([
albu.RandomFog(fog_coef_lower=0.1, fog_coef_upper=0.15, p=0.1),
albu.RandomShadow(p=0.1),
albu.RandomBrightness(limit=0.3, p=0.2),
albu.RandomRain(slant_lower=0,
slant_upper=8,
drop_length=0,
blur_value=4,
brightness_coefficient=0.8,
rain_type='heavy',
p=0.1),
albu.RandomSunFlare(p=0.2),
]),
albu.OneOf([
albu.RGBShift(p=0.1),
albu.HueSaturationValue(p=0.3),
]),
albu.OneOf([
albu.HorizontalFlip(p=0.5),
albu.RandomSizedCrop(min_max_height=(720, 1380),
height=1380,
width=720,
interpolation=cv2.INTER_AREA)
],
p=0.2)
]
return albu.Compose(train_transform,
bbox_params={
'format': 'coco',
'min_area': min_area,
'min_visibility': min_visibility,
'label_fields': ['category_id']
})
def hard_color_augmentations():
return A.Compose([
A.RandomBrightnessContrast(brightness_limit=0.3,
contrast_limit=0.3,
brightness_by_max=True),
A.RandomGamma(gamma_limit=(90, 110)),
A.OneOf(
[A.NoOp(),
A.MultiplicativeNoise(),
A.GaussNoise(),
A.ISONoise()]),
A.OneOf([A.RGBShift(), A.HueSaturationValue(),
A.NoOp()]),
A.RandomFog(fog_coef_lower=0.05, fog_coef_upper=0.3),
])
def setup_pipeline(self, dict_transform):
tranform_list = []
if 'shadow' in dict_transform:
tranform_list.append(
A.RandomShadow(shadow_roi=(0, 0.5, 1, 1),
num_shadows_upper=1,
p=0.2))
if 'scale' in dict_transform:
tranform_list.append(
A.RandomScale(scale_limit=float(dict_transform['scale'])))
if 'rotate' in dict_transform:
tranform_list.append(
A.Rotate(limit=float(dict_transform['rotate']), p=0.8))
if 'shift' in dict_transform:
tranform_list.append(
A.ShiftScaleRotate(shift_limit=float(dict_transform['shift']),
scale_limit=0.0,
rotate_limit=0,
interpolation=1,
border_mode=4,
p=0.8))
if 'brightness' in dict_transform:
tranform_list.append(
A.RandomBrightness(limit=float(dict_transform['brightness']),
p=0.8))
if 'contrast' in dict_transform:
tranform_list.append(
A.RandomContrast(limit=float(dict_transform['contrast']),
p=0.8))
if 'motion_blur' in dict_transform:
tranform_list.append(A.MotionBlur(p=0.5, blur_limit=7))
if 'fog' in dict_transform:
tranform_list.append(
A.RandomFog(fog_coef_lower=0.0,
fog_coef_upper=float(dict_transform['fog']),
alpha_coef=0.05,
p=0.7))
if 'rain' in dict_transform:
tranform_list.append(
A.RandomRain(brightness_coefficient=0.95,
drop_width=1,
blur_value=1,
p=0.7))
if 'occlusion' in dict_transform:
tranform_list.append(
A.CoarseDropout(max_holes=5, max_height=8, max_width=8, p=0.5))
self.transform = A.Compose(tranform_list)
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 __init__(self, cfg):
self.cfg = cfg
self.data = self.prepare()
self.mean = self.cfg.mean
self.std = self.cfg.std
self.normal_transform = A.Compose([
A.Resize(384, 288, p=1.0),
A.HorizontalFlip(p=0.5),
A.Normalize(p=1.0, mean=self.mean, std=self.std)
])
self.augment_transform = A.Compose([
A.Resize(384, 288, p=1.0),
A.HorizontalFlip(p=0.7),
A.GaussNoise(p=0.5),
A.ShiftScaleRotate(shift_limit=0.1,
scale_limit=0.25,
rotate_limit=20,
p=0.6,
border_mode=0),
A.OneOf([
A.CLAHE(p=0.5),
A.Compose([
A.RandomBrightness(limit=0.5, p=0.6),
A.RandomContrast(limit=0.4, p=0.6),
A.RandomGamma(p=0.6),
])
],
p=0.65),
A.OneOf([
A.HueSaturationValue(10, 20, 10, p=1.0),
A.RGBShift(p=1.0),
A.Emboss(p=1.0),
],
p=0.5),
A.RandomFog(fog_coef_lower=0.3, fog_coef_upper=0.3, p=0.3),
A.OneOf([
A.Perspective(p=1.0, scale=(0.05, 0.1)),
A.GridDistortion(p=1.0, distort_limit=0.25, border_mode=0),
A.OpticalDistortion(
p=1.0, shift_limit=0.1, distort_limit=0.1, border_mode=0)
],
p=0.65),
A.Normalize(p=1.0, mean=self.mean, std=self.std),
])
def __init__(self):
self.policy = A.Compose([
A.OneOf([
A.Rotate(180),
A.Flip(),
], p=0.3),
A.ShiftScaleRotate(shift_limit=0.2, scale_limit=0.5, rotate_limit=0, p=0.2),
A.OneOf([
A.CoarseDropout(max_holes=16, max_height=16, max_width=16, p=0.3),
A.GridDropout(ratio=0.3, p=0.3),
]),
A.OneOf([
A.ElasticTransform(sigma=10, alpha_affine=25, p=0.3),
A.RandomFog(fog_coef_lower=0.2, fog_coef_upper=0.7, p=0.2),
], p=0.2),
A.OneOf([
A.IAAAdditiveGaussianNoise(),
A.GaussNoise(),
A.ISONoise()
], p=0.2),
A.OneOf([
A.MotionBlur(p=.3),
A.MedianBlur(blur_limit=5, p=0.3),
A.Blur(blur_limit=5, p=0.3),
A.GaussianBlur(p=0.3)
], p=0.2),
A.OneOf([
A.ChannelShuffle(p=.3),
A.HueSaturationValue(p=0.3),
A.ToGray(p=0.3),
A.ChannelDropout(p=0.3),
A.InvertImg(p=0.1)
], p=0.2),
A.OneOf([
A.OpticalDistortion(p=0.3),
A.GridDistortion(p=.2),
A.IAAPiecewiseAffine(p=0.3),
], p=0.2),
A.OneOf([
A.CLAHE(clip_limit=2),
A.IAASharpen(),
A.IAAEmboss(),
], p=0.2),
A.RandomBrightnessContrast(brightness_limit=0.3, contrast_limit=0.3, p=0.3),
A.Solarize(p=0.2),
])
def noise_transforms(p=0.5):
"""
Applies GaussNoise or RandomFog random with a probability p.
Args:
p (float, optional): probability. Defaults to 0.5.
Returns:
albumentation transforms: transforms.
"""
return albu.OneOf(
[
albu.GaussNoise(var_limit=(1.0, 50.0), always_apply=True),
albu.RandomFog(
fog_coef_lower=0.01, fog_coef_upper=0.25, always_apply=True),
],
p=p,
)
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 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 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 augmentation_function(n_augment: int) -> A.core.composition.Compose:
"""Performs data augmentation on images with realistic parameters
Returns a Albumentation composition function"""
proba_transfo = 1 / n_augment
def random_proba(proba_transfo: float) -> float:
return max(0.4, min(1, proba_transfo + (2 * np.random.random() - 1) / 3))
return A.Compose(
[
A.HorizontalFlip(p=1),
A.Rotate(limit=25, p=random_proba(proba_transfo)),
A.RandomBrightness(limit=0.1, p=random_proba(proba_transfo)),
A.RandomSnow(brightness_coeff=0.95, snow_point_lower=0.1,
snow_point_upper=0.3, p=random_proba(proba_transfo)),
A.RandomFog(fog_coef_lower=0.1, fog_coef_upper=0.4,
alpha_coef=0.1, p=random_proba(proba_transfo)),
],
p=1.0,
bbox_params=A.BboxParams(format='yolo', label_fields=['category_ids'])
)
def augment(self, img, mask, do_affine_transform = True):
if do_affine_transform:
afine_transform = A.Compose([
A.HorizontalFlip(p=0.4),
A.OneOf([
A.GridDistortion(interpolation=cv2.INTER_NEAREST, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
A.ElasticTransform(interpolation=cv2.INTER_NEAREST, alpha_affine=10, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
A.ShiftScaleRotate(interpolation=cv2.INTER_NEAREST, shift_limit=0.03, rotate_limit=4, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
A.OpticalDistortion(interpolation=cv2.INTER_NEAREST, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0)
], p=0.6),
], additional_targets={'mask': 'image'})
afine_transformed = afine_transform(image=img, mask=mask)
img = afine_transformed["image"]
mask = afine_transformed["mask"]
transform = A.Compose([
A.IAAAdditiveGaussianNoise(p=0.05),
A.OneOf([
A.IAASharpen(p=1.0),
A.Blur(blur_limit=3, p=1.0),
] , p=0.5),
A.OneOf([
A.RandomBrightnessContrast(p=1.0),
A.HueSaturationValue(p=1.0),
A.RandomGamma(p=1.0),
], p=0.5),
A.OneOf([
A.RandomFog(p=1.0),
A.RandomRain(p=1.0),
A.RandomShadow(p=1.0),
A.RandomSnow(p=1.0),
A.RandomSunFlare(p=1.0)
], p=0.05),
])
transformed = transform(image=img)
img = transformed["image"]
return img, mask
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 get_transforms(aug_type: str):
"""
Data augmentation 객체 생성
Args:
aug_type(str) : augmentation타입 지정
Returns :
list :: train, validation, test데이터 셋에 대한 transform
"""
# TODO: Normalize
if False:
pass
else:
norm_mean = (0, 0, 0)
norm_std = (1, 1, 1)
if aug_type == 'no':
train_transform = A.Compose(
[A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()])
val_transform = A.Compose(
[A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()])
test_transform = A.Compose(
[A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()])
elif aug_type == 'dev':
pass
elif aug_type == 'final':
train_transform = A.Compose([
A.HorizontalFlip(p=0.5),
A.RandomResizedCrop(512,
512,
p=0.8,
scale=(0.7, 1.0),
ratio=(0.5, 1.5)),
A.Rotate(limit=30, p=0.8),
A.Cutout(num_holes=4, max_h_size=16, max_w_size=16, p=0.8),
A.ElasticTransform(alpha=40, p=0.8),
A.CLAHE(clip_limit=3.0, p=0.8),
A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()
])
val_transform = A.Compose(
[A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()])
test_transform = A.Compose(
[A.Normalize(mean=norm_mean, std=norm_std),
ToTensorV2()])
# TODO: 아래 코드 정리 중
elif aug_type == 'basic':
train_transform = A.Compose([
A.Normalize(mean=(0.46009655, 0.43957878, 0.41827092),
std=(0.2108204, 0.20766491, 0.21656131),
max_pixel_value=255.0,
p=1.0),
ToTensorV2()
])
val_transform = A.Compose([
A.Normalize(mean=(0.46009655, 0.43957878, 0.41827092),
std=(0.2108204, 0.20766491, 0.21656131),
max_pixel_value=255.0,
p=1.0),
ToTensorV2()
])
test_transform = A.Compose([ToTensorV2()])
elif aug_type == "aug1":
train_transform = A.Compose([
A.OneOf([
A.ElasticTransform(alpha=120,
sigma=120 * 0.05,
alpha_affine=120 * 0.03,
p=0.5),
A.GridDistortion(p=1.0),
A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=1)
],
p=0.8),
A.VerticalFlip(p=0.5),
A.RandomBrightnessContrast(p=0.8),
A.RandomGamma(p=0.8),
A.RandomRotate90(p=0.5),
ToTensorV2()
])
val_transform = A.Compose([ToTensorV2()])
test_transform = A.Compose([ToTensorV2()])
elif aug_type == "aug2":
train_transform = A.Compose([
A.OneOf([
A.RandomShadow(p=1),
A.RandomSunFlare(num_flare_circles_lower=1,
num_flare_circles_upper=5,
src_radius=250,
p=1),
A.RandomRain(p=1),
A.RandomSnow(brightness_coeff=1.5, p=1),
A.RandomFog(
fog_coef_lower=0.8, fog_coef_upper=1, alpha_coef=0.08, p=1)
],
p=0.8),
ToTensorV2()
])
val_transform = A.Compose([ToTensorV2()])
test_transform = A.Compose([ToTensorV2()])
elif aug_type == "aug3":
train_transform = A.Compose([
A.OneOf([
A.HueSaturationValue(hue_shift_limit=0.2,
sat_shift_limit=0.2,
val_shift_limit=0.2,
p=1),
A.MultiplicativeNoise(
multiplier=(0.9, 1.1), per_channel=True, p=1),
A.RGBShift(r_shift_limit=0.1,
g_shift_limit=0.1,
b_shift_limit=0.1,
p=1),
A.ChannelShuffle(0.05)
],
p=0.8),
ToTensorV2()
])
val_transform = A.Compose([ToTensorV2()])
test_transform = A.Compose([ToTensorV2()])
elif aug_type == "aug4":
train_transform = A.Compose([
A.HorizontalFlip(p=0.5),
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
ToTensorV2()
])
val_transform = A.Compose([ToTensorV2()])
test_transform = A.Compose([ToTensorV2()])
elif aug_type == "aug5":
train_transform = A.Compose([
A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
#A.ElasticTransform(p=1, alpha=120, sigma=120 * 0.05, alpha_affine=120 * 0.03),
#A.GridDistortion(p=1.0),
#A.OpticalDistortion(distort_limit=2, shift_limit=0.5, p=1)
#A.VerticalFlip(p=0.5),
#A.RandomBrightnessContrast(p=0.8),
#A.RandomGamma(p=0.8),
#A.RandomRotate90(p=0.5),
#A.MultiplicativeNoise(multiplier=(0.9, 1.1), per_channel=True, p=0.8),
ToTensorV2()
])
val_transform = A.Compose([ToTensorV2()])
test_transform = A.Compose([ToTensorV2()])
return train_transform, val_transform, test_transform
def train_function(gpu, world_size, node_rank, gpus):
import torch.multiprocessing
torch.multiprocessing.set_sharing_strategy('file_system')
torch.manual_seed(25)
np.random.seed(25)
rank = node_rank * gpus + gpu
dist.init_process_group(
backend='nccl',
init_method='env://',
world_size=world_size,
rank=rank
)
width_size = 512
batch_size = 32
accumulation_step = 5
device = torch.device("cuda:{}".format(gpu) if torch.cuda.is_available() else "cpu")
if rank == 0:
wandb.init(project='inception_v3', group=wandb.util.generate_id())
wandb.config.width_size = width_size
wandb.config.aspect_rate = 1
wandb.config.batch_size = batch_size
wandb.config.accumulation_step = accumulation_step
shutil.rmtree('tensorboard_runs', ignore_errors=True)
writer = SummaryWriter(log_dir='tensorboard_runs', filename_suffix=str(time.time()))
ranzcr_df = pd.read_csv('train_folds.csv')
ranzcr_train_df = ranzcr_df[ranzcr_df['fold'] != 1]
chestx_df = pd.read_csv('chestx_pseudolabeled_data_lazy_balancing.csv')
train_image_transforms = alb.Compose([
alb.ImageCompression(quality_lower=65, p=0.5),
alb.HorizontalFlip(p=0.5),
alb.CLAHE(p=0.5),
alb.OneOf([
alb.GridDistortion(
num_steps=8,
distort_limit=0.5,
p=1.0
),
alb.OpticalDistortion(
distort_limit=0.5,
shift_limit=0.5,
p=1.0,
),
alb.ElasticTransform(alpha=3, p=1.0)],
p=0.7
),
alb.RandomResizedCrop(
height=width_size,
width=width_size,
scale=(0.8, 1.2),
p=0.7
),
alb.RGBShift(p=0.5),
alb.RandomSunFlare(p=0.5),
alb.RandomFog(p=0.5),
alb.RandomBrightnessContrast(p=0.5),
alb.HueSaturationValue(
hue_shift_limit=20,
sat_shift_limit=20,
val_shift_limit=20,
p=0.5
),
alb.ShiftScaleRotate(shift_limit=0.025, scale_limit=0.1, rotate_limit=20, p=0.5),
alb.CoarseDropout(
max_holes=12,
min_holes=6,
max_height=int(width_size / 6),
max_width=int(width_size / 6),
min_height=int(width_size / 6),
min_width=int(width_size / 20),
p=0.5
),
alb.IAAAdditiveGaussianNoise(loc=0, scale=(2.5500000000000003, 12.75), per_channel=False, p=0.5),
alb.IAAAffine(scale=1.0, translate_percent=None, translate_px=None, rotate=0.0, shear=0.0, order=1, cval=0,
mode='reflect', p=0.5),
alb.IAAAffine(rotate=90., p=0.5),
alb.IAAAffine(rotate=180., p=0.5),
alb.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
ToTensorV2()
])
train_set = NoisyStudentDataset(ranzcr_train_df, chestx_df, train_image_transforms,
'../ranzcr/train', '../data', width_size=width_size)
train_sampler = DistributedSampler(train_set, num_replicas=world_size, rank=rank, shuffle=True)
train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=False, num_workers=4, sampler=train_sampler)
ranzcr_valid_df = ranzcr_df[ranzcr_df['fold'] == 1]
valid_image_transforms = alb.Compose([
alb.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
ToTensorV2()
])
valid_set = ImageDataset(ranzcr_valid_df, valid_image_transforms, '../ranzcr/train', width_size=width_size)
valid_loader = DataLoader(valid_set, batch_size=batch_size, num_workers=4, pin_memory=False, drop_last=False)
# ranzcr_valid_df = ranzcr_df[ranzcr_df['fold'] == 1]
# valid_image_transforms = alb.Compose([
# alb.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
# ToTensorV2()
# ])
# valid_set = ImageDataset(ranzcr_valid_df, valid_image_transforms, '../ranzcr/train', width_size=width_size)
# valid_sampler = DistributedSampler(valid_set, num_replicas=world_size, rank=rank)
# valid_loader = DataLoader(valid_set, batch_size=batch_size, num_workers=4, sampler=valid_sampler)
checkpoints_dir_name = 'inception_v3_noisy_student_{}'.format(width_size)
os.makedirs(checkpoints_dir_name, exist_ok=True)
# model = EfficientNetNoisyStudent(11, pretrained_backbone=True,
# mixed_precision=True, model_name='tf_efficientnet_b7_ns')
model = Inception(11, pretrained_backbone=True, mixed_precision=False, model_name='inception_v3')
model = SyncBatchNorm.convert_sync_batchnorm(model)
model.to(device)
model = DistributedDataParallel(model, device_ids=[gpu])
# class_weights = [354.625, 23.73913043478261, 2.777105767812362, 110.32608695652173,
# 52.679245283018865, 9.152656621728786, 4.7851333032083145,
# 8.437891632878731, 2.4620064899945917, 0.4034751151063363, 31.534942820838626]
class_names = ['ETT - Abnormal', 'ETT - Borderline', 'ETT - Normal',
'NGT - Abnormal', 'NGT - Borderline', 'NGT - Incompletely Imaged', 'NGT - Normal',
'CVC - Abnormal', 'CVC - Borderline', 'CVC - Normal', 'Swan Ganz Catheter Present']
scaler = GradScaler()
criterion = torch.nn.BCEWithLogitsLoss()
lr_start = 1e-4
lr_end = 1e-6
weight_decay = 0
epoch_num = 20
if rank == 0:
wandb.config.model_name = checkpoints_dir_name
wandb.config.lr_start = lr_start
wandb.config.lr_end = lr_end
wandb.config.weight_decay = weight_decay
wandb.config.epoch_num = epoch_num
wandb.config.optimizer = 'adam'
wandb.config.scheduler = 'CosineAnnealingLR'
wandb.config.is_loss_weights = 'no'
optimizer = Adam(model.parameters(), lr=lr_start, weight_decay=weight_decay)
scheduler = CosineAnnealingLR(optimizer, T_max=epoch_num, eta_min=lr_end, last_epoch=-1)
max_val_auc = 0
for epoch in range(epoch_num):
train_loss, train_avg_auc, train_auc, train_rocs, train_data_pr, train_duration = one_epoch_train(
model, train_loader, optimizer, criterion, device, scaler,
iters_to_accumulate=accumulation_step, clip_grads=False)
scheduler.step()
if rank == 0:
val_loss, val_avg_auc, val_auc, val_rocs, val_data_pr, val_duration = eval_model(
model, valid_loader, device, criterion, scaler)
wandb.log({'train_loss': train_loss, 'val_loss': val_loss,
'train_auc': train_avg_auc, 'val_auc': val_avg_auc, 'epoch': epoch})
for class_name, auc1, auc2 in zip(class_names, train_auc, val_auc):
wandb.log({'{} train auc'.format(class_name): auc1,
'{} val auc'.format(class_name): auc2, 'epoch': epoch})
if val_avg_auc > max_val_auc:
max_val_auc = val_avg_auc
wandb.run.summary["best_accuracy"] = val_avg_auc
print('EPOCH %d:\tTRAIN [duration %.3f sec, loss: %.3f, avg auc: %.3f]\t\t'
'VAL [duration %.3f sec, loss: %.3f, avg auc: %.3f]\tCurrent time %s' %
(epoch + 1, train_duration, train_loss, train_avg_auc,
val_duration, val_loss, val_avg_auc, str(datetime.now(timezone('Europe/Moscow')))))
torch.save(model.module.state_dict(),
os.path.join(checkpoints_dir_name, '{}_epoch{}_val_auc{}_loss{}_train_auc{}_loss{}.pth'.format(
checkpoints_dir_name, epoch + 1, round(val_avg_auc, 3), round(val_loss, 3),
round(train_avg_auc, 3), round(train_loss, 3))))
if rank == 0:
wandb.finish()
建议把所有加噪声的方法加入此类
"""
def __init__(self, transforms):
self.transforms = transforms
def __call__(self, img):
n = len(self.transforms)
idx = np.random.randint(0, n - 1)
t = self.transforms[idx]
return t(image=img)
rand_weather = OneOf([
am.RandomRain(),
am.RandomFog(),
am.RandomSnow(),
]) # 策略2，随机天气，自然界可能随机发生一些天气现象，增强模型对天气现象的抗干扰能力，参考链接https://github.com/UjjwalSaxena/Automold--Road-Augmentation-Library
sharpOrBlur = OneOf([
am.GaussianBlur(), # 高斯模糊在mnist中我已经提供了一个自己实现的版本，这里选择调库
am.IAASharpen(),
])
noiser = OneOf([
GaussNoise(),
SaltAndPepper(),
])
strategy = [eraser, rand_weather, sharpOrBlur, noiser]
def __init__(self, args, main_dir='/opt/ml/', mode='train'):
self.postfix = args.postfix
self.main_dir = Path(main_dir)
self.data_dir = self.main_dir / 'input/data'
self.image_folder = self.data_dir / mode / 'cropped_images'
self.meta_dir = self.data_dir / mode / str(mode + '.csv')
self.gridshuffle = True if args.gridshuffle == 1 else False
self.mixed_precision = True if args.mixed_precision == 1 else False
self.n_fold = args.n_fold
self.s_epoch = args.s_epoch
self.t_epoch = args.t_epoch
self.weight_path = None
self.weighed_sampler = True if args.weighted_sampler == 1 else False
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
self.nosiy_elimination = True if args.nosiy_elimination == 1 else False
# criterion
self.clipping = True if "nfnet" in args.model_type else False
self.crit = args.crit
self.arcface_crit = args.arcface_crit
self.focal_type = args.focal_type
self.cls_weight = True if args.cls_weight == 1 else False
self.focal_gamma = 5.0
# optimizer
self.optim = args.optim
self.lr = args.lr
self.weight_decay = args.decay
# scheduler
self.sched_type = args.sched_type
self.sched_T_0 = args.T_max if args.T_max != 0 else self.t_epoch
self.eta_min = args.eta_min
# model
self.cls_num = 18
self.backbone_name = args.model_type
self.checkpoint = self.main_dir / 'checkpoints' / str(self.backbone_name + "_" + args.postfix)
if not os.path.exists(self.checkpoint):
os.makedirs(self.checkpoint, exist_ok=True)
self.backbone_pretrained = True if mode == 'train' else False
self.embed_size = args.embed_size
self.pool = args.pool
self.p_trainable = True
self.neck = args.neck
self.multi_dropout = True if args.multi_dropout == 1 else False
self.multi_dropout_num = 16
self.multi_dropout_prob = 0.2
# pseudo label
self.pseudo_label = True if args.pseudo_label == 1 else False
self.pseudo_label_data = self.main_dir / 'submission' / args.pseudo_label_path
# logging
self.log_interval = 50
self.log_dir = self.main_dir / 'logs'
if not os.path.exists(self.log_dir):
os.makedirs(self.log_dir, exist_ok=True)
self.log_dir = self.log_dir / (self.backbone_name + "_" + args.postfix + '.txt')
self.mean = [0.56019358, 0.52410121, 0.501457]
self.std = [0.23318603, 0.24300033, 0.24567522]
# transforms
self.trn_tfms = A.Compose([
A.Resize(384, 288, p=1.0),
A.HorizontalFlip(p=0.7),
A.GaussNoise(p=0.5),
A.ShiftScaleRotate(shift_limit=0.1, scale_limit=0.25, rotate_limit=20, p=0.6, border_mode=0),
A.OneOf([
A.CLAHE(p=0.5),
A.Compose([
A.RandomBrightness(limit=0.5, p=0.6),
A.RandomContrast(limit=0.4, p=0.6),
A.RandomGamma(p=0.6),
])
], p=0.65),
A.OneOf([
A.HueSaturationValue(10, 20, 10, p=1.0),
A.RGBShift(p=1.0),
A.Emboss(p=1.0),
], p=0.5),
A.RandomFog(fog_coef_lower=0.3, fog_coef_upper=0.3, p=0.3),
A.OneOf([
A.Perspective(p=1.0, scale=(0.05, 0.1)),
A.GridDistortion(p=1.0, distort_limit=0.25, border_mode=0),
A.OpticalDistortion(p=1.0, shift_limit=0.1, distort_limit=0.1, border_mode=0)
], p=0.65),
A.Normalize(p=1.0, mean=self.mean, std=self.std),
])
self.val_tfms = A.Compose([
A.Resize(384, 288),
A.Normalize(p=1.0, mean=self.mean, std=self.std),
])
def get_config(runner,
raw_uri,
processed_uri,
root_uri,
test=False,
external_model=False,
external_loss=False,
augment=False):
debug = False
train_scene_info = get_scene_info(join(processed_uri, 'train-scenes.csv'))
val_scene_info = get_scene_info(join(processed_uri, 'val-scenes.csv'))
log_tensorboard = True
run_tensorboard = True
class_config = ClassConfig(names=['no_building', 'building'])
if test:
debug = True
train_scene_info = train_scene_info[0:1]
val_scene_info = val_scene_info[0:1]
def make_scene(scene_info):
(raster_uri, label_uri) = scene_info
raster_uri = join(raw_uri, raster_uri)
label_uri = join(processed_uri, label_uri)
aoi_uri = join(raw_uri, aoi_path)
if test:
crop_uri = join(processed_uri, 'crops',
os.path.basename(raster_uri))
label_crop_uri = join(processed_uri, 'crops',
os.path.basename(label_uri))
save_image_crop(raster_uri,
crop_uri,
label_uri=label_uri,
label_crop_uri=label_crop_uri,
size=600,
min_features=20,
class_config=class_config)
raster_uri = crop_uri
label_uri = label_crop_uri
id = os.path.splitext(os.path.basename(raster_uri))[0]
raster_source = RasterioSourceConfig(channel_order=[0, 1, 2],
uris=[raster_uri])
label_source = ChipClassificationLabelSourceConfig(
vector_source=GeoJSONVectorSourceConfig(uri=label_uri,
default_class_id=1,
ignore_crs_field=True),
ioa_thresh=0.5,
use_intersection_over_cell=False,
pick_min_class_id=False,
background_class_id=0,
infer_cells=True)
return SceneConfig(id=id,
raster_source=raster_source,
label_source=label_source,
aoi_uris=[aoi_uri])
chip_sz = 200
train_scenes = [make_scene(info) for info in train_scene_info]
val_scenes = [make_scene(info) for info in val_scene_info]
dataset = DatasetConfig(class_config=class_config,
train_scenes=train_scenes,
validation_scenes=val_scenes)
if external_model:
model = ClassificationModelConfig(external_def=ExternalModuleConfig(
github_repo='lukemelas/EfficientNet-PyTorch',
# uri='s3://raster-vision-ahassan/models/EfficientNet-PyTorch.zip',
name='efficient_net',
entrypoint='efficientnet_b0',
force_reload=False,
entrypoint_kwargs={
'num_classes': len(class_config.names),
'pretrained': 'imagenet'
}))
else:
model = ClassificationModelConfig(backbone=Backbone.resnet50)
if external_loss:
external_loss_def = ExternalModuleConfig(
github_repo='AdeelH/pytorch-multi-class-focal-loss',
name='focal_loss',
entrypoint='focal_loss',
force_reload=False,
entrypoint_kwargs={
'alpha': [.75, .25],
'gamma': 2
})
else:
external_loss_def = None
solver = SolverConfig(lr=1e-4,
num_epochs=20,
test_num_epochs=4,
batch_sz=32,
one_cycle=True,
external_loss_def=external_loss_def)
if augment:
mu = np.array((0.485, 0.456, 0.406))
std = np.array((0.229, 0.224, 0.225))
aug_transform = A.Compose([
A.Flip(),
A.Transpose(),
A.RandomRotate90(),
A.ShiftScaleRotate(),
A.OneOf([
A.ChannelShuffle(),
A.CLAHE(),
A.FancyPCA(),
A.HueSaturationValue(),
A.RGBShift(),
A.ToGray(),
A.ToSepia(),
]),
A.OneOf([
A.RandomBrightness(),
A.RandomGamma(),
]),
A.OneOf([
A.GaussNoise(),
A.ISONoise(),
A.RandomFog(),
]),
A.OneOf([
A.Blur(),
A.MotionBlur(),
A.ImageCompression(),
A.Downscale(),
]),
A.CoarseDropout(max_height=32, max_width=32, max_holes=5)
])
base_transform = A.Normalize(mean=mu.tolist(), std=std.tolist())
plot_transform = A.Normalize(mean=(-mu / std).tolist(),
std=(1 / std).tolist(),
max_pixel_value=1.)
else:
aug_transform = None
base_transform = None
plot_transform = None
backend = PyTorchChipClassificationConfig(
model=model,
solver=solver,
log_tensorboard=log_tensorboard,
run_tensorboard=run_tensorboard,
test_mode=test,
base_transform=A.to_dict(base_transform),
aug_transform=A.to_dict(aug_transform),
plot_options=PlotOptions(transform=A.to_dict(plot_transform)))
config = ChipClassificationConfig(root_uri=root_uri,
dataset=dataset,
backend=backend,
train_chip_sz=chip_sz,
predict_chip_sz=chip_sz)
return config
A.RandomRotate90(),
A.ShiftScaleRotate(scale_limit = 0.2, rotate_limit = 20, p = 0.9, border_mode=cv2.BORDER_REFLECT),
A.OneOf([
A.IAAPiecewiseAffine(),
A.GridDistortion(),
A.OpticalDistortion(),
], p = 1),
A.OneOf([
A.RandomBrightnessContrast(),
A.HueSaturationValue(),
A.CLAHE()
], p = 1),
A.OneOf([
A.RandomRain(),
A.RandomSnow(),
A.RandomFog()
], p = 1),
A.Resize(size, size),
A.Normalize()
], p = 1)
valid_trans = A.Compose([
A.Resize(size, size),
A.Normalize()
])
split = StratifiedKFold(n_folds, shuffle = True, random_state = seed)
model = None
for fold in range(fold, n_folds):
METRIC = 0.
logger(f"\n*******START: seed {seed}, fold {fold}*******")
def augment(self,
img1, bbox1 = [], keypoints1 = [], filtered_objs1 = [],
img0 = None, bbox0 = [], keypoints0 = [], filtered_objs0 = [],
do_img_aug = True, do_affine_aug = False, img1_to_img0 = False
):
if img0 is None:
img0 = img1.copy()
replay_img_aug = []
keypoints = keypoints1 + keypoints0
bboxes = bbox1 + bbox0
# Afine augmentations
# --------------------------------------
if do_affine_aug:
afine_transform = A.ReplayCompose(
[
A.HorizontalFlip(p=0.4),
A.OneOf([
# A.GridDistortion(interpolation=cv2.INTER_NEAREST, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
# A.ElasticTransform(interpolation=cv2.INTER_NEAREST, alpha_affine=10, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
A.ShiftScaleRotate(interpolation=cv2.INTER_NEAREST, shift_limit=0.035, rotate_limit=5, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
# A.OpticalDistortion(interpolation=cv2.INTER_NEAREST, border_mode=cv2.BORDER_CONSTANT, value=0, mask_value=0, p=1.0),
], p=0.65),
],
additional_targets={"img0": "image"},
keypoint_params=A.KeypointParams(format="xy", remove_invisible=False),
bbox_params=A.BboxParams(format='coco', label_fields=["bbox_ids"]), # coco format: [x-min, y-min, width, height]
)
transformed = afine_transform(image=img1, img0=img0, keypoints=keypoints, bboxes=bboxes, bbox_ids=np.arange(len(bboxes)))
replay_img_aug.append(transformed["replay"])
img1 = transformed["image"]
img0 = transformed["img0"]
transformed_keypoints = transformed['keypoints']
transformed_bboxes = transformed['bboxes']
bbox_ids = transformed["bbox_ids"]
# it can happend that bounding boxes are removed, we have to account for that
# and also remove the objects and keypoints in question
idx_offset = len(bbox1)
keypoints1 = []
keypoints0 = []
bbox1 = []
bbox0 = []
tmp_filtered_objs1 = []
tmp_filtered_objs0 = []
for i, bbox_id in enumerate(bbox_ids):
kp_idx_start = bbox_id * 8
kp_idx_end = bbox_id * 8 + 8
if bbox_id < idx_offset:
tmp_filtered_objs1.append(filtered_objs1[i])
bbox1.append(transformed_bboxes[i])
keypoints1 += transformed_keypoints[kp_idx_start:kp_idx_end]
else:
tmp_filtered_objs0.append(filtered_objs0[i])
bbox0.append(transformed_bboxes[i])
keypoints0 += transformed_keypoints[kp_idx_start:kp_idx_end]
filtered_objs1 = tmp_filtered_objs1
filtered_objs0 = tmp_filtered_objs0
# Translate img0 + bbox0/keypoints0 for single track centernet
# --------------------------------------
if img1_to_img0:
transform = A.Compose(
[A.ShiftScaleRotate(shift_limit=0.1, scale_limit=0.15, rotate_limit=0, always_apply=True, border_mode=cv2.BORDER_CONSTANT)],
keypoint_params=A.KeypointParams(format="xy", remove_invisible=False),
bbox_params=A.BboxParams(format='coco', label_fields=[])
)
keypoints = list(np.array(piped_params["gt_2d_info"])[:,:2] * self.params.R)
transformed = transform(image=img1, keypoints=keypoints1, bboxes=bboxes1)
img0 = transformed["image"]
bboxes0 = transformed['bboxes']
keypoints0 = transformed['keypoints']
# Augmentation for images
# --------------------------------------
if do_img_aug:
transform = A.ReplayCompose([
A.IAAAdditiveGaussianNoise(p=0.02),
A.OneOf([
A.IAASharpen(p=1.0),
A.Blur(blur_limit=3, p=1.0),
] , p=0.5),
A.OneOf([
A.RandomBrightnessContrast(p=1.0),
A.HueSaturationValue(p=1.0),
A.RandomGamma(p=1.0),
], p=0.6),
A.OneOf([
A.RandomFog(p=1.0),
A.RandomRain(p=1.0),
A.RandomShadow(p=1.0),
A.RandomSnow(p=1.0)
], p=0.02),
], additional_targets={"img0": "image"})
transformed = transform(image=img1, img0=img0)
img1 = transformed["image"]
img0 = transformed["img0"]
return img1, bbox1, keypoints1, filtered_objs1, img0, bbox0, keypoints0, filtered_objs0, replay_img_aug
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
def getdata(img, count, disp=False):
#randomly transform the template image to generate sample data
#need to add a new function that generates data by registering real -
#input images with template and thus generating transform matrix
xtrain = []
ytrain = []
d = 20
pts1 = np.float32([[0, 0], [width, 0], [width, height]])
'''
cv2.circle(img, tuple(pts1[0]), 5, (0, 0, 255), -1)
cv2.circle(img, tuple(pts1[1]), 5, (0, 0, 255), -1)
cv2.circle(img, tuple(pts1[2]), 5, (0, 0, 255), -1)
cv2.imshow("Image", img)
'''
aug = A.Compose(
[
A.RandomBrightnessContrast(p=.5),
A.RandomGamma(p=.5),
A.RandomRain(
brightness_coefficient=0.9, drop_width=1, blur_value=5, p=.5),
A.RandomShadow(num_shadows_lower=1,
num_shadows_upper=1,
shadow_dimension=5,
shadow_roi=(0, 0.5, 1, 1),
p=.5),
A.RandomFog(
fog_coef_lower=0.1, fog_coef_upper=0.5, alpha_coef=0.1, p=.5),
#A.RandomSunFlare(flare_roi=(0, 0, 0.2, 0.2), angle_lower=0.2, p=.5),
#A.CLAHE(p=1),
A.HueSaturationValue(hue_shift_limit=3,
sat_shift_limit=50,
val_shift_limit=50,
p=.5),
],
p=.8)
for i in range(count):
d0 = random.sample(range(-d, d), 2)
d1 = random.sample(range(-d, d), 2)
d2 = random.sample(range(-d, d), 2)
#import pdb;pdb.set_trace()
pts2 = np.float32([pts1[0] + d0, pts1[1] + d1, pts1[2] + d2])
matrix = cv2.getAffineTransform(pts1, pts2)
result = cv2.warpAffine(img, matrix, (width, height))
#augmentations
result = aug(image=result)['image']
#import pdb;pdb.set_trace()
matrix = cv2.invertAffineTransform(matrix)
matrix = matrix.flatten()
xtrain.append(result)
ytrain.append(matrix)
if disp == True:
cv2.imshow("Affine transformation", result)
cv2.waitKey(30)
cv2.destroyAllWindows()
xtrain = np.array(xtrain, dtype=np.float32)
ytrain = np.array(ytrain, dtype=np.float32)
return (xtrain / 255.0, ytrain)
import albumentations as A
transform_fn = A.Compose([
A.OneOf([
A.IAAAdditiveGaussianNoise(),
A.GaussNoise(),
], p=.2),
A.OneOf([
A.MotionBlur(p=.2),
A.MedianBlur(blur_limit=3, p=.2),
A.Blur(blur_limit=3, p=.2),
], p=.2),
A.OneOf([
A.CLAHE(clip_limit=2),
A.IAASharpen(),
A.IAAEmboss(),
A.RandomBrightnessContrast(),
], p=0.2),
A.OneOf([
A.ChannelShuffle(),
A.HueSaturationValue(),
A.RGBShift(),
A.ToSepia(),
], p=0.2),
A.OneOf([
A.RandomSunFlare(flare_roi=(0, 0, 1., 1.), src_radius=100, p=.1),
A.RandomFog(fog_coef_lower=0.2, fog_coef_upper=0.3, p=.1),
A.RandomShadow(p=.1),
A.RandomRain(p=.1, blur_value=1),
], p=0.1)
])
def __call__(self, data):
rgb, thermal, depth, audio, label, id = data
rgb = rgb.astype(np.float32)
height, width, _ = rgb.shape
albumentations_transform_pixel = {
'Blur': albumentations.Blur(),
#'CLAHE':albumentations.CLAHE(),
'ChannelDropout': albumentations.ChannelDropout(),
'ChannelShuffle': albumentations.ChannelShuffle(),
'CoarseDropout': albumentations.CoarseDropout(),
#'Equalize':albumentations.Equalize(),
#'FancyPCA':albumentations.FancyPCA(),
'GaussNoise': albumentations.GaussNoise(),
'GaussianBlur': albumentations.GaussianBlur(),
#'GlassBlur':albumentations.GlassBlur(),
'HueSaturationValue': albumentations.HueSaturationValue(),
'IAAAdditiveGaussianNoise':
albumentations.IAAAdditiveGaussianNoise(),
#'ISONoise':albumentations.ISONoise(),
'RGBShift': albumentations.RGBShift(),
'RandomBrightnessContrast':
albumentations.RandomBrightnessContrast(),
'RandomFog': albumentations.RandomFog(),
#'RandomGamma':albumentations.RandomGamma(),
'RandomRain': albumentations.RandomRain(),
'RandomShadow': albumentations.RandomShadow(),
'RandomSnow': albumentations.RandomSnow(),
'RandomSunFlare': albumentations.RandomSunFlare(),
'Solarize': albumentations.Solarize(),
}
albumentations_transform_bbox = {
#'HorizontalFlip':albumentations.HorizontalFlip(),
#'VerticalFlip':albumentations.VerticalFlip(),
#'CenterCrop':albumentations.CenterCrop(height=height-10, width=width-10, p=0.5),
#'RandomCropNearBBox':albumentations.RandomCropNearBBox(p=0.5),
#'Crop':albumentations.Crop(x_min=10, y_min =10, y_max=height-10, x_max=width-10, p=0.5),
#'ElasticTransform':albumentations.ElasticTransform(),
#'ShiftScaleRotate':albumentations.ShiftScaleRotate(),
}
transform = np.random.choice(
['None'] + list(albumentations_transform_pixel.keys()) +
list(albumentations_transform_bbox.keys()))
if transform in albumentations_transform_pixel:
aug = albumentations.Compose(
[albumentations_transform_pixel[transform]],
bbox_params={
'format': 'pascal_voc',
'label_fields': ['labels']
})
try:
annots = np.array(annots).astype(np.float32)
aug_result = aug(image=rgb,
bboxes=annots[:, :4],
labels=annots[:, 4])
rgb = aug_result['image']
annots = np.hstack([
aug_result['bboxes'],
np.array(aug_result['labels']).reshape(-1, 1)
])
except Exception as e:
print(
f"transform={transform} aug_result['bboxes']={aug_result['bboxes']} aug_result['labels']={aug_result['labels']}"
)
raise Exception(e)
elif transform in albumentations_transform_bbox:
aug = albumentations.Compose(
[albumentations_transform_bbox[transform]],
bbox_params={
'format': 'pascal_voc',
'label_fields': ['labels']
})
try:
annots = np.array(annots).astype(np.float32)
aug_result = aug(image=rgb,
bboxes=annots[:, :4],
labels=annots[:, 4])
rgb = aug_result['image']
label = np.hstack([
aug_result['bboxes'],
np.array(aug_result['labels']).reshape(-1, 1)
])
except Exception as e:
print(
f"transform={transform} aug_result['bboxes']={aug_result['bboxes']} aug_result['labels']={aug_result['labels']}"
)
raise Exception(e)
return rgb, thermal, depth, audio, label, id