FLOPs can not reveal real latency of model, but MACs can?

sovrasov / flops-counter.pytorch

Flops counter for convolutional networks in pytorch framework

MIT License

2.82k stars 307 forks source link

FLOPs can not reveal real latency of model, but MACs can? #70

Open lucasjinreal opened 3 years ago

lucasjinreal commented 3 years ago

from the last table only saw MACs and params, is that FLOPs doesn't needed a record? While some paper still report their FLOPs/G for comparasion.

rongzhao-zhang commented 3 years ago

Either FLOPs or MACs are all theoretical things. The real latency is also highly dependent on the hardware structures/drivers, etc., which is extremely hard to predict in practice.

lucasjinreal commented 3 years ago

Oh, I mean why not also add a FLOPs in column in common used model in README table.

sovrasov commented 3 years ago

@jinfagang see #16