How to dump Custom Dataset into ImageFolder readable format?

vision
1

I have a CustomDataset that subclasses torch.utils.data.Dataset.
Due to some third-party library specs, I have an unusual situation where I need the data in CustomDataset to be stored in a directory in an ImageFolder readable format.

What’s the easiest/cleanest way to achieve this?
Many thanks

2

If you need to store the data in the “folder layout” expected by ImageFolder, you could either move the data to the root and subfolders or create symbolic links instead.

3

Can you explain what do you mean by move the data to the root and subfolders? The CustomDataset contains custom data that I did not load from a folder but created from scratch

4

In that case I don’t understand the question correctly.
Could you explain this part a bit more:

If you are creating the data from scratch, then you might want to save it to the expected folder structure.

5

Yes, I am creating the data from scratch and have it in the form of a pytorch Dataset and I was wondering if there’s a simple way to “export” it into an ImageFolder compatible format.
I imagine I can do this “manually” myself, but I wanted to know if there’s some existing functionality that can easily dump a Dataset into the desired format, as the ImageFolder format is common.

6

Thanks for clarifying. No, I don’t think there is an automatic way to export the images.

1 Like