Post-processing (Clustering) Slow
rayhou0710 opened this issue · comments
Thank you for publishing the code!
I ran the test using your pretrained model on the car class but found out that the avg post-processing time is about 391 ms per image on a Titan Xp GPU which is much slower than the number reported in the paper. May I ask if I missed anything? Thanks!
Hi rayhou, it is indeed correct that the clustering is slower than mentioned in the paper. This is because this repo includes a python version of the clustering, and not the c++ variant which we used for timing. Given our license agreement, we only release the 'research' version and not the optimized production version. However, given the python version, I think it is quite easy to convert to c++, you should then definitely see an improvement.
Thank you for the reply!