Computer Vision using the CIFAR-10 dataset

[christophergaughan]

Spurred on by the ComputerVision learning we did with the FashionMNIST dataset, which was grey-scale (i.e. 1 color channel), is it possible to follow along with our code to use a 3 color channel dataset (like CIFAR-10)? If so, the issues I have run into so far have been the color channel order where I begin with a shape of the data being [32, 3, 32, 32] (batch num, color channels, H, W). I have noticed I have had to use the permute() in order to get the color channel order correct (i.e. when plotting you need to switch from formats [3, 32, 32] -> [32, 32, 3]. Will this type of project become too difficult when training a model? Do you know of any resources out there to help me do this type of project? I want to stick mostly with the code we wrote with just the aforementioned adjustments. Any help on this would be appreciated. Thanks for any answers.

[christophergaughan]

I'll answer my own question, I hadn't watched the PyTorch_custom_Datasets, in there I have my answers. However, I'm glad I tried it on my own so I could see the utility of the transform method.