def test_data_normalized(self):
"""Checks if the image data provided by this dataset is in [0.0,1.0]
given the ToTensor() transformation."""
dataset = regression_dataset.RegressionDataset(_DATASET_PATH,
transforms.ToTensor())
for img_data, _ in dataset:
self.assertTrue(_is_normalized(img_data.numpy().flatten()))
def test_order_of_entries(self):
"""dataset[k] returns the kth (img_data, score) tuple. Here, we just
test equality on scores. See the dataset.json file in _DATASET_PATH for
scores."""
dataset = regression_dataset.RegressionDataset(_DATASET_PATH)
self.assertEqual(dataset[0][1], 0.998)
self.assertEqual(dataset[1][1], 0.734)
self.assertEqual(dataset[2][1], 0.343)
self.assertEqual(dataset[3][1], 0.123)
def _loader(dataset_path, batch_size):
'''Returns a DataLoader that emits (image_data, score) tuples. Data is not
shuffled and crops are always center cropped.'''
_IMAGENET_MEAN = [0.485, 0.456, 0.406]
_IMAGENET_STD = [0.229, 0.224, 0.225]
transform = transforms.Compose([
transforms.CenterCrop([224, 224]),
transforms.ToTensor(),
transforms.Normalize(_IMAGENET_MEAN, _IMAGENET_STD),
])
dataset = regression_dataset.RegressionDataset(dataset_path, transform)
data_loader = torch.utils.data.DataLoader(
dataset, batch_size=batch_size, shuffle=False, num_workers=4)
return data_loader
def _loader(dataset_path, batch_size):
'''Returns a DataLoader that emits (image_data, score) tuples. Dataset is
randomly shuffled and randomly cropped.'''
_IMAGENET_MEAN = [0.485, 0.456, 0.406]
_IMAGENET_STD = [0.229, 0.224, 0.225]
transform = transforms.Compose([
# TODO (carlo): Figure out a sensible way to do data augmentations here.
transforms.RandomCrop([224, 224]),
transforms.ToTensor(),
transforms.Normalize(_IMAGENET_MEAN, _IMAGENET_STD),
])
dataset = regression_dataset.RegressionDataset(dataset_path, transform)
data_loader = torch.utils.data.DataLoader(dataset,
batch_size=batch_size,
shuffle=True,
num_workers=4)
return data_loader
def test_length(self):
dataset = regression_dataset.RegressionDataset(_DATASET_PATH)
self.assertEqual(len(dataset), 4)