Questions regarding the optimal placement of FDConv layers

[Javis-101]

Is FDConv primarily recommended for use in shallow networks? When applied to layers with high channel counts (e.g., [2, 768, 128, 128]), the computational cost becomes prohibitively high!

[Linwei-Chen]

Thanks for your interest! FDConv is also recommended for shallow networks; feel free to try it. The code has recently been updated for improved GPU efficiency.