Open fffffq99 opened 1 day ago
fbxiang
commented
23 hours ago OptiX SDK does not need to be installed.
Since you are building from source, I would assume there is something special about your system that cannot support x86 manylinux? Otherwise maybe you can reproduce on the released version.
Another option is to try oidn instead of OptiX. My experience is that oidn is often both higher quality and faster than optix. OptiX just had more features that we do not use anyway.
fffffq99
commented
14 hours ago Thank you. Now my problem has been solved.
At first, I ran the rt_mat.py in the documentation.
It was specified to use optix there, but since I couldn't use optix, I manually changed it to oidn. But only this result can be produced:
So I thought it was because oidn was not good enough, so I tried to figure out how to use optix. After your reply, I rechecked the code and found that this is a problem with the code:
rgb = im.fromarray((rgb * 255).astype(np.uint8))
Due to the absence of clip, this issue occurred, revise to:
rgb = im.fromarray((rgb * 255).clip(0, 255).astype(np.uint8))
This way, I can reproduce pretty results.
So, I think you should modify the rt_mat.py in the documentation to prevent others from experiencing this problem as well.
System:
Describe the bug When I set
sapien.render.set_ray_tracing_denoiser("optix"), an error is displayed : [2024-11-04 11:00:15.108] [svulkan2] [error] OptiX Error: OPTIX_ERROR_INVALID_VALUE [2024-11-04 11:00:15.108] [svulkan2] [error] OptiX Error: OPTIX_ERROR_INVALID_VALUE [2024-11-04 11:00:15.110] [svulkan2] [error] Failed to denoise OptiX Error: OPTIX_ERROR_INVALID_VALUEBut if I set 'none' or 'oidn', it can work. I suspect that I may not have installed the NVIDIA OptiX SDK correctly, but I don't know how to install it to make SAPIEN work properly. For example, should I install it on the host or inside Docker? How should I set up a dynamic link library? I need some help, can you help me? Look forward to your reply.