Lexie88rus

# Compose an augmentation pipeline
aug_pipeline = A.Compose([
    A.ShiftScaleRotate(),
    A.RGBShift(),
    A.Blur(),
    A.GaussNoise()
],p=1)

# Load the mask
mask = imageio.imread('mask1.png')

Lexie88rus

# import Albumentations package
import albumentations as A

# Initialize an augmentation
gaus_noise = A.GaussNoise() # gaussian noise

# Apply augmentation
img_gaus = gaus_noise(image = image)

# Access the augmented image by the 'image' key

Lexie88rus

# Import PyTorch
import torchvision.transforms.functional as TF
from torch.utils.data import Dataset, DataLoader

# Define the augmentations
AUG_TRAIN = aug_pipeline # use our pipeline as train augmentations

# Define the demo dataset
class DogDataset(Dataset):
    '''

Lexie88rus

# Define an augmentation pipeline
aug_pipeline = iaa.Sequential([
    iaa.Sometimes(0.5, iaa.GaussianBlur((0, 3.0))), # apply Gaussian blur with a sigma between 0 and 3 to 50% of the images
    # apply one of the augmentations: Dropout or CoarseDropout
    iaa.OneOf([
        iaa.Dropout((0.01, 0.1), per_channel=0.5), # randomly remove up to 10% of the pixels
        iaa.CoarseDropout((0.03, 0.15), size_percent=(0.02, 0.05), per_channel=0.2),
    ]),
    # apply from 0 to 3 of the augmentations from the list
    iaa.SomeOf((0, 3),[

Lexie88rus

# Plot the initial and the augmented images with bounding boxes
# using helpers from imgaug package
side_by_side = np.hstack([
    bbs.draw_on_image(image, size=2), # blend the original image with bounding box
    bbs_aug.draw_on_image(image_aug, size=2) # blend the augmented image with bounding box
])

# Plot with matplotlib imshow()
fig, ax = plt.subplots(figsize=(10, 7))
ax.axis('off')

Lexie88rus

# Import bounding boxes from imgaug
from imgaug.augmentables.bbs import BoundingBox, BoundingBoxesOnImage

# Initialize the bounding box for the original image
# using helpers from imgaug package
bbs = BoundingBoxesOnImage([
    BoundingBox(x1=1, x2=980, y1=9, y2=535)
], shape=image.shape)

# Define a simple augmentations pipeline for the image with bounding box

Lexie88rus

# Import matplotlib for visualization
from matplotlib import pyplot as plt

# Visualize original image, augmented image, and segmentation map
side_by_side = np.hstack([
    segmap.draw_on_image(image), # show blend of original image and segmentation map
    segmap_aug.draw_on_image(image_aug),  # show blend of augmented image and segmentation map
    segmap_aug.draw()  # show only the augmented segmentation map
])

Lexie88rus

# Import PIL
from PIL import Image

# Import segmentation maps from imgaug
from imgaug.augmentables.segmaps import SegmentationMapOnImage

# Open image with mask
pil_mask = Image.open('../input/mask1.png')

# Convert mask to binary map

Lexie88rus

# import imageio to open images
import imageio
# import augmenters from imgaug
from imgaug import augmenters as iaa

# use imageio library to read the image (alternatively you can use OpenCV cv2.imread() function)
image = imageio.imread('image1.jpg')

# initialize the augmenter
rotate = iaa.Affine(rotate=(-25, 25)) # rotate image

Lexie88rus

def silu_inplace_2(input):
    '''
    Example of implementation of in-place SiLU activation function using torch.sigmoid_
    https://arxiv.org/pdf/1606.08415.pdf
    '''
    result = input.clone()
    torch.sigmoid_(input)
    input *= result
    return input