I wonder how much the flops is when num_blocks=[1,1,1]?Is F-conv more computationally efficient than others in experiments?

[MugeYan]

Noticed that the comparison experiments in paper seem to conceal the computational cost metrics, which is kinda unfair for comparisons, I wonder how much the FLOPs is when num_blocks=[1,1,1]?Is F-conv more computationally efficient than others in experiments?Would be much appreciate to get positive feedbacks.