Replace pycuda with torch extensions

[nexuslrf]

Hi,

I made a fork of this repo (https://github.com/nexuslrf/pointnerf/tree/master/models/neural_points/c_ext) and replaced all pycuda modules with torch extension, the CUDA kernels are almost the same as before. I verified the modifications by running w_n360/lego.sh & dtu_test_inf/inftest_scan1.sh.

For those who are still struggling with pycuda, you can give it try.

I am willing to create a pull request for this, but it might need more verifications from the community.

[RuixiangXue]

Hi @nexuslrf , I also meet the same question about pycuda, but I have no idea about how to try your thought. Can u share your code about this file?

[nexuslrf]

Hi @RuixiangXue , you can replace/add all the changed files in this commit:
master...nexuslrf:pointnerf:master

After that, run the cmd shown in https://github.com/nexuslrf/pointnerf/blob/master/models/neural_points/c_ext/README.md

[Xharlie]

hi, we have updated the pytorch cuda extension 5 months ago, is your extension similar?

[CTouch]

goooooood job!

[xiaoc57]

If you have any doubts， you can replace the orgin models folder directly. This work is helpful for me.

[lhp121]

After running the command python setup.py build_ext --inplace, can you directly run or test the code, such as bash dev_scripts/w_n360/lego_test.sh?@nexuslrf