Hi everyone,
I was wondering if it is possible to obtain the original name or the full path of the image after using torch.utils.data.random_split(). Achieving this would allow to save train and test split images with their original names to local directories. Thanks.
random_split
wraps the Dataset
into Subsets
as seen here. The used indices can be access via subset_dataset.indices
and could be used to get the image paths.
It depends what kind of Dataset
you are using, but e.g. ImageFolder
should contains the paths in its dataset.samples
attribute.
Thanks a lot @ptrblck for the reply.