You can implement an idiomatic CNN in PyTorch or TF and compare their performance to this implementation. It's a perfectly reasonable comparison.