Availability of (target_)?transform(s)? arguments in torchvision.datasets

torchvision_datasets_transforms.md

This is an overview over the usage of the availability of the transform, target_transform, and transforms arguments for all datasets from torchvision.datasets.

Dataset	transform	target_transform	transforms
Caltech101	x	x
Caltech256	x	x
CelebA	x	x
CIFAR10	x	x
CIFAR100	x	x
Cityscapes	x	x	x
CocoCaptions	x	x	x
CocoDetection	x	x	x
FakeData	x	x
Flickr8k	x	x
Flickr30k	x	x
DatasetFolder	x	x
ImageFolder	x	x
HMDB51	x
ImageNet	x	x
Kinetics400	x
LSUN	x	x
MNIST	x	x
FashionMNIST	x	x
KMNIST	x	x
EMNIST	x	x
QMNIST	x	x
Omniglot	x	x
Phototour	x
Places365	x	x
SBDataset			x
SBU	x	x
SEMEION	x	x
STL10	x	x
SVHN	x	x
UCF101	x
USPS	x	x
VOCSegmentation	x	x	x
VOCDetection	x	x	x
WIDERFace	x	x

Most image datasets accept the transform / target_transform pair. There are however some exceptions:

• Cityscapes, CocoCaptions, CocoDetection, VOCSegmentation, and VOCDetection which additionally accept joint transforms
• Phototour only accepts a transform
• SBDataset only accepts joint transforms

The video datasets HMDB51, Kinetics400, and UCF101 uniformly only accept a transform