ma-xu/pointMLP-pytorch

How to obtain the speed in Table 2?

inspirelt opened this issue · 2 comments

Hi,

thanks for the excellent job. I'm confusing on how to obtain the speed in Table 2. For pointMLPElite, I set the test batch_size = 16, and it takes 5s to process the whole 2468 test samples(~494samples/s). With test batch_size=1, it takes about 24s -> ~103samples/s. Both results are far different from the 176samples/s in Table 2. Would you mind explaining the speedtest operation in details?

ma-xu commented

Hi @inspirelt ,

Thanks for your interest.

I tested all methods using the first option.
I just uploaded the tested time results here pointmlp time , point-elite time. The testing times are 22s/ 14s (See out.txt file),respectively. Hence, the speed would be 112samples/s and 176samples/s. Please ignore the model name because it was the name of our developing version.

I guess the difference comes from the machines. Note we are tested on one Tesla V100-
pcie GPU
and 4 cores AMD EPYC 7351 2.60GHz CPU.

Please let me know if any further questions.

Thanks for your reply.