Load .png into PyTorch with different folders as labels

Hi, everyone!

I’m pretty new on PyTorch, and until now I’ve done several tutorials in which I think they download the data from torchvision but at the moment I never loaded my own data.

I am trying to do a classification task with three labels, and I have several png’s distributed in three different folders depending on the label.

I have the folder ‘dataset’ in the same folder where the .py file I’m working on is, and the path from here is ‘/dataset/train/each label’.

So there are two questions:

How do I get to the png files, upload them and transform them to tensors? It seems that it’s not looking inside the same folder where my PyTorch file is…

And second, once I get to the files, how do I label the files in each folder? Usually, as long as I made models, label and sample are in the same file…

The code I have for now is this:

import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms

train_path = './dataset/train/Covid'

train_dataset = torchvision.datasets.ImageFolder(

It doesn’t even find the directory…

Thanks in advance!