This repository reproduced a paper. 2014 Flattened convolutional neural networks for feedforward acceleration
There is a lua and cuda code version https://github.com/jhjin/flattened-cnn
It is difficult to understand, so I developed an easy to understand version base on pyTorch.
My version has a disadvantage, too slow. This is because img2col function is implemented by for loop.
In the future, I prepare implement img2col by cuda code.
reference: