data_transformations.py

import torchvision.transforms as transforms

__all__ = ['tensor_transform']

# custom class for data transformation can also be written

# Use for the SSL dataset
tensor_transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.5032, 0.4746, 0.4275), (0.2276, 0.2228, 0.2265))
])

# Use for the SSL dataset
resnet_input_transform = transforms.Compose([
	transforms.RandomAffine((-5,5), translate=(0.1,0.1), scale=(0.9, 1.1), shear=None, resample=False, fillcolor=0),
    transforms.Resize((96, 96)),
    transforms.ToTensor(),
    transforms.Normalize((0.5032, 0.4746, 0.4275), (0.2276, 0.2228, 0.2265))
])

# use only for the CIFAR Dataset
cifar10_input_transform = transforms.Compose([
	transforms.RandomAffine((-5,5), translate=(0.1,0.1), scale=(0.9, 1.1), shear=None, resample=False, fillcolor=0),
    transforms.Resize((48, 48)),
    transforms.ToTensor(),
    transforms.Normalize((0.4925, 0.4828, 0.4464), (0.2024, 0.1998, 0.2006))
])

# SSL Data
# transforms.Normalize((0.5032, 0.4746, 0.4275), (0.2276, 0.2228, 0.2265))

# for CIFAR data
# transforms.Normalize((0.4925, 0.4828, 0.4464), (0.2024, 0.1998, 0.2006))