About the experiment in the paper

[870231652]

I have a question for you. I did an experiment using your method mentioned in the paper, which is stacking 10 pure convolutions. However, I found that the results were not ideal, and the FLOPS of Pconv were even lower than those of GPU on the CPU. I used a 3060 graphics card. Can you help me take a look? Is there any problem with the code? Thank you very much.