def get_dataset(name,
split='train',
transform=None,
target_transform=None,
download=True,
datasets_path=__DATASETS_DEFAULT_PATH):
train = (split == 'train')
root = os.path.join(datasets_path, name)
if name == 'cifar10_whitened':
x = load_lua('/media/SSD/Datasets/cifar10/cifar10_whitened.t7')
if train:
return dataset.TensorDataset([
x['trainData']['data'],
(x['trainData']['labels'] - 5.5).sign()
])
else:
return dataset.TensorDataset([
x['testData']['data'], (x['testData']['labels'] - 5.5).sign()
])
if name == 'tinyImagenet':
if train:
return datasets.ImageFolder(
root='/home/ehoffer/Datasets/ImageNet/tiny',
transform=transform,
target_transform=target_transform)
# x = load_lua('/home/ehoffer/Datasets/ImageNet/tinyImageNet.t7')
# return dataset.TensorDataset(
# [x['data'], (x['label'].float()-500.5).sign()])
else:
return dataset.TensorDataset(
[torch.rand(100, 3, 64, 64).float(),
torch.rand(100).float()])
elif name == 'cifar10':
return datasets.CIFAR10(root=root,
train=train,
transform=transform,
target_transform=target_transform,
download=download)
elif name == 'cifar100':
return datasets.CIFAR100(root=root,
train=train,
transform=transform,
target_transform=target_transform,
download=download)
elif name == 'mnist':
return datasets.MNIST(root=root,
train=train,
transform=transform,
target_transform=target_transform,
download=download)
elif name == 'imagenet':
if train:
root = os.path.join(root, 'train')
else:
root = os.path.join(root, 'val')
return datasets.ImageFolder(root=root,
transform=transform,
target_transform=target_transform)
def testTensorDataset(self):
# dict input
data = {
# 'input': torch.range(0,7),
'input': np.arange(0, 8),
'target': np.arange(0, 8),
}
d = dataset.TensorDataset(data)
self.assertEqual(len(d), 8)
self.assertEqual(d[2], {'input': 2, 'target': 2})
# tensor input
a = torch.randn(8)
d = dataset.TensorDataset(a)
self.assertEqual(len(a), len(d))
self.assertEqual(a[1], d[1])
# list of tensors input
d = dataset.TensorDataset([a])
self.assertEqual(len(a), len(d))
self.assertEqual(a[1], d[1][0])
def testShuffleDataset(self):
tbl = dataset.TensorDataset(np.asarray([0, 1, 2, 3, 4]))
d = dataset.ShuffleDataset(tbl)
self.assertEqual(len(d), 5)
def testResampleDataset(self):
tbl = dataset.TensorDataset(np.asarray([0, 1, 2]))
d = dataset.ResampleDataset(tbl, lambda dataset, i: i % 2)
self.assertEqual(len(d), 3)
self.assertEqual(d[0], 0)
self.assertEqual(d[2], 0)