How can I create images for each batch using Pytorch?

Question

I want to make a binary classifier that classifies the following:

Class 1. Some images that I already have.

Class 2. Some images that I create from a function, using the images of class 1.

The problem is that instead of pre-creating the two classes, and then loading them, to speed up the process I would like the class 2 images to be created for each batch.

Any ideas on how I can tackle the problem? If I use the DataLoader as usual, I have to enter the images of both classes directly, but if I still don't have the images of the second class I don't know how to do it.

Thanks.

Do you want to save the images of class 2 after processing or only preprocessing them and then train the model? — Hamzah Al-Qadasi
– Hamzah Al-Qadasi, Commented Mar 8, 2022 at 19:02
Well a simple way of doing it instead of datasets and loaders is a simple stack to create 1 batch via the torch.stack function — Edwin Cheong
– Edwin Cheong, Commented Mar 9, 2022 at 3:08

aretor · Accepted Answer · 2022-03-08 19:30:17Z

1

You can tackle the problem in at least two ways.

(Preferred) You create a custom Dataset class, AugDset, such that AugDset.__len__() returns 2 * len(real_dset), and when idx > len(imgset), AugDset.__getitem__(idx) generates the synthetic image from real_dset(idx).
You create your custom collate_fn function, to be passed to DataLoader that, given a batch, it augments it with your synthetic generated images.

answered Mar 8, 2022 at 19:30

aretor

2,5695 gold badges25 silver badges42 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

How can I create images for each batch using Pytorch?

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related