Is the count_conv2d for FLOPs? about pytorch-opcounter HOT 7 CLOSED

Umm, I don't get what you mean. Can you explain more detailly?

from pytorch-opcounter.

I understand that one multiplication or one addition can be seen as one FLOP.
I wonder what is mean the multiply-add = 1 in your code. How does change when multiply-add = 2?

And can you know me how can I calculate MACs ?

from pytorch-opcounter.

In some cases, a multiplication is treated as 2 additions. So I put a multiply_adds constant to ease the modification.

MACs = 2 * FLOPs, you may convert it by yourself. https://en.wikipedia.org/wiki/Multiply%E2%80%93accumulate_operation

from pytorch-opcounter.

Generally, MACs consider multiplication and addition as one operation, as far as I know. Are the FLOPs two operation in that case?

from pytorch-opcounter.

I think so.

from pytorch-opcounter.

In this equation MACs = 2 * FLOPs, If I want to get MACs, do I have to multiply FLOPs by 2?
MACs can be bigger than FLOPs according to the equation. But I think the MACs should be smaller than FLOPs.
Does that equation mean multiply-add = 2?
I refer to this site.
https://machinethink.net/blog/how-fast-is-my-model/

from pytorch-opcounter.

Do you plan to make bilinear interpolate layer?

from pytorch-opcounter.