(May be) BUGs in paper: wrong Gflops

[Sixkplus]

Thank you for your nice work.

I started to reproduce swiftnet in tensorflow and check the layers one by one carefully. However, the tf.profile always gives the GFLOPs over 200:

I even calculated the flops of each conv layers, which also sums to ~200 GFlops

I find the reason today. The torchstat gives the following FLOPS number.

This tool MAY give MADD and FLOPs WRONGLY. The correct GFlops should be around 200 and 104 should be GMadd Swall0w/torchstat/issues/12