ywangeq/PointSeg

visualization code of test_demo.py is very time consuming

Closed this issue · 1 comments

kxhit commented

Thanks for your great work!
I successfully trained the model and run the test_demo.py to visualize the result. But I noticed that the GPU(test on TITAN XP) takes around 0.01s, while read points costs 0.20s!!! and pub to ros costs 0.04s!
It finally makes the visualization in rviz performs poorly. I think it is unreasonable that rospy is so time consuming. Is there any way to deal with that? Thanks!

ywangeq commented

Sorry for late response, I don't receive the information from my email which connected this account.
For your questions, the test_demo.py shows the process from raw data to suitable input(64,512,5), which I only use for loop to insert points one by one. I think you can 'import numba as nb ' and use ‘@jit’ to speed that for loop part.