2d dct FLOPs computing method

[Tianhao Fu]

Hi, I noticed that in your paper you computed FCAnet model FLOPs.

I wonder how do you compute the FLOPs of 2d dct? Could you provide your formula or code?

Thanks!

[peterzpy]

We compute the weight of 2d dct during initialization. The dct kernel is fixed and has no extra computation cost in the training phase.

[Tianhao Fu]

We compute the weight of 2d dct during initialization. The dct kernel is fixed and has no extra computation cost in the training phase.

Thanks.
Leave the paper aside. Do you know how to compute the FLOPs of 2d dct?

If you can provide me some idea or code or fomula about that, I will be thanks so much:) 👍 @peterzpy

[cfzd]

@TianhaoFu
The realization of 2d dct is actually element product. For a image with a size h x w and a dct basis with a size of h x w, the re would be total hw times of multiplication.

[cfzd]

The FLOPs is calculated by: https://github.com/sovrasov/flops-counter.pytorch

[Tianhao Fu]

Thanks