transform = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
# Initialize CNN
K = 500 # total number of exemplars
icarl = iCaRLNet(2048, 3)
icarl.cuda()
for s in range(0, total_classes, num_classes):
# Load Datasets
print("Loading training examples for classes", range(s, s + num_classes))
# train_set = iCIFAR10(root='./data',
# train=True,
# classes=range(s,s+num_classes),
# download=True,
# transform=transform_test)
train_set = mnist(train=True,
classes=range(s, s + num_classes),
transform=transform_test)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=128,
transform = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
# Initialize CNN
K = 2000 # total number of exemplars
icarl = iCaRLNet(2048, 10)
icarl.cuda()
for s in range(0, total_classes, num_classes):
# Load Datasets
print("Loading training examples for classes", range(s, s + num_classes))
train_set = iCIFAR100(root='./data',
train=True,
classes=range(s, s + num_classes),
download=True,
transform=transform_test)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=64,
shuffle=True,
num_workers=0)