training stuck at Renderer

[YESAndy]

Hi,

Thank you very much for releasing your work! I was trying to rerun the gs training but stuck for a long time (~30 mins) when running

im, radius, _, = Renderer(raster_settings=curr_data['cam'])(**rendervar) from get_loss function in train.py

I suspected that it may be because of the installation configuration problem about CUDA. So I tried to reinstall the diff-gaussian-rasterization-w-depth package with CUDA_HOME=/path/to/cuda. But the problem still existed.

I checked the source code of the Renderer which is generated by python binding codes from C++. I have zero idea about how to debug with C++ ;)

Do you have any suggestion about the reason why it stuck? Thank you!

[YESAndy]

problem solved. it is because the mismatch between nvidia driver and torch version:)