Hi all!
For the Neural Networks tutorial in the Deep Learning with PyTorch: A 60 Minute Blitz , we implement LeNet that performs well on the MNIST dataset. The tutorial assumes that the input images are size 32x32, where as when you load the MNIST dataset from torchvision, the image sizes are 28x28.
The only change I did was change self.fc1(1655,120) to self.fc1(1644,120), and now I can train on the MNIST dataset.
Would it make sense to add some code to upsample the images, or change fc1 to handle the 28x28 dimensions? Either way, I think it would be helpful to tweak the tutorial so someone can run it on the actual MNIST dataset, or just have a disclaimer of the difference a person would need to consider.