Dataset/Dataloader to Load Batches per Class

vision
1

My dataset is in the format

root/
    1/
        1.png
        2.png
        3.png
        ...
    2/
        1.png
        2.png
        3.png
        ...
    ...

I need to be able to load batches containing images only from a specific class, for each class.
Eg. (1st batch are 8 images from root/1, 2nd batch are 8 images from root/2, so on)

Is there any way to do so without having multiple datasets, each for a particular class?

Any help is appreciated.

2

If I understand you correctly, the ImageFolder class does just that:

https://pytorch.org/docs/stable/torchvision/datasets.html#imagefolder

3

But is it possible to specify which class to sample from?

4

Not with the ImageFolder object. But you could perhaps create an ImageFolder object for each class to be able to sample independently from each class.