Datasets — Torchvision 0.22 documentation (original) (raw)

Torchvision provides many built-in datasets in the torchvision.datasetsmodule, as well as utility classes for building your own datasets.

Built-in datasets¶

All datasets are subclasses of torch.utils.data.Dataseti.e, they have __getitem__ and __len__ methods implemented. Hence, they can all be passed to a torch.utils.data.DataLoaderwhich can load multiple samples in parallel using torch.multiprocessing workers. For example:

imagenet_data = torchvision.datasets.ImageNet('path/to/imagenet_root/') data_loader = torch.utils.data.DataLoader(imagenet_data, batch_size=4, shuffle=True, num_workers=args.nThreads)

All the datasets have almost similar API. They all have two common arguments:transform and target_transform to transform the input and target respectively. You can also create your own datasets using the provided base classes.

Warning

When a dataset object is created with download=True, the files are first downloaded and extracted in the root directory. This download logic is not multi-process safe, so it may lead to conflicts / race conditions if it is run within a distributed setting. In distributed mode, we recommend creating a dummy dataset object to trigger the download logic before setting up distributed mode.

Image classification¶

Image detection or segmentation¶

Optical Flow¶

Stereo Matching¶

Image pairs¶

Image captioning¶

Video classification¶

Video prediction¶

Base classes for custom datasets¶

Transforms v2¶