def test_dataloader(self,
batch_size,
num_images_per_class=-1,
transforms=None):
"""
Uses the validation split of imagenet2012 for testing
Args:
batch_size: the batch size
num_images_per_class: how many images per class to test on
transforms: the transforms
"""
transforms = self.val_transform(
) if self.test_transforms is None else self.test_transforms
dataset = UnlabeledImagenet(self.data_dir,
num_imgs_per_class=num_images_per_class,
meta_dir=self.meta_dir,
split='test',
transform=transforms)
loader = DataLoader(dataset,
batch_size=batch_size,
shuffle=False,
num_workers=self.num_workers,
drop_last=True,
pin_memory=True)
return loader
def imagenet_val(dataset_root, nb_classes):
train_transform = amdim_transforms.TransformsImageNet128()
dataset = UnlabeledImagenet(dataset_root,
nb_classes=nb_classes,
split='val',
transform=train_transform)
return dataset
def train_dataloader(self, batch_size, transforms=None):
"""
Uses the train split of imagenet2012 and puts away a portion of it for the validation split
Args:
batch_size: the batch size
transforms: the transforms
"""
if transforms is None:
transforms = self.train_transform()
dataset = UnlabeledImagenet(self.data_dir,
num_imgs_per_class=-1,
meta_root=self.meta_root,
split='train',
transform=transforms)
loader = DataLoader(
dataset,
batch_size=batch_size,
shuffle=True,
num_workers=self.num_workers,
drop_last=True,
pin_memory=True
)
return loader
def imagenet_train(dataset_root, nb_classes, patch_size, patch_overlap):
train_transform = amdim_transforms.TransformsImageNet128Patches(
patch_size=patch_size, overlap=patch_overlap)
dataset = UnlabeledImagenet(dataset_root,
nb_classes=nb_classes,
split='train',
transform=train_transform)
return dataset
def imagenet(dataset_root, nb_classes, split: str = 'train'):
assert split in ('train', 'val')
dataset = UnlabeledImagenet(
dataset_root,
nb_classes=nb_classes,
split=split,
transform=amdim_transforms.AMDIMTrainTransformsImageNet128(),
)
return dataset
def imagenet(dataset_root, nb_classes, patch_size, patch_overlap, split: str = 'train'):
assert split in ('train', 'val')
train_transform = amdim_transforms.TransformsImageNet128Patches(
patch_size=patch_size,
overlap=patch_overlap
)
dataset = UnlabeledImagenet(
dataset_root,
nb_classes=nb_classes,
split=split,
transform=train_transform,
)
return dataset
def val_dataloader(self,
batch_size,
num_images_per_class=50,
add_normalize=False):
transforms = self._default_transforms(
) if self.val_transforms is None else self.val_transforms
dataset = UnlabeledImagenet(
self.data_dir,
num_imgs_per_class_val_split=num_images_per_class,
meta_dir=self.meta_dir,
split='val',
transform=transforms)
loader = DataLoader(dataset,
batch_size=batch_size,
shuffle=False,
num_workers=self.num_workers,
pin_memory=True)
return loader
def test_dataloader(self):
"""
Uses the validation split of imagenet2012 for testing
"""
transforms = self.val_transform(
) if self.test_transforms is None else self.test_transforms
dataset = UnlabeledImagenet(self.data_dir,
num_imgs_per_class=-1,
meta_dir=self.meta_dir,
split='test',
transform=transforms)
loader = DataLoader(dataset,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers,
drop_last=True,
pin_memory=True)
return loader
def train_dataloader(self):
"""
Uses the train split of imagenet2012 and puts away a portion of it for the validation split
"""
transforms = self.train_transform(
) if self.train_transforms is None else self.train_transforms
dataset = UnlabeledImagenet(self.data_dir,
num_imgs_per_class=-1,
meta_dir=self.meta_dir,
split='train',
transform=transforms)
loader = DataLoader(dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.num_workers,
drop_last=True,
pin_memory=True)
return loader
def train_dataloader(self,
batch_size,
num_images_per_class=-1,
transforms=None,
add_normalize=False):
if transforms is None:
transforms = self._default_transforms()
dataset = UnlabeledImagenet(self.data_dir,
num_imgs_per_class=num_images_per_class,
meta_root=self.meta_root,
split='train',
transform=transforms)
loader = DataLoader(dataset,
batch_size=batch_size,
shuffle=True,
num_workers=self.num_workers,
drop_last=True,
pin_memory=True)
return loader
def val_dataloader(self):
"""
Uses the part of the train split of imagenet2012 that was not used for training via `num_imgs_per_val_class`
Args:
batch_size: the batch size
transforms: the transforms
"""
transforms = self.train_transform(
) if self.val_transforms is None else self.val_transforms
dataset = UnlabeledImagenet(
self.data_dir,
num_imgs_per_class_val_split=self.num_imgs_per_val_class,
meta_dir=self.meta_dir,
split='val',
transform=transforms)
loader = DataLoader(dataset,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers,
pin_memory=True)
return loader
