42yeah/GSRast

error

Opened this issue · 8 comments

jaccen commented

Bad reader to path: shaders/pointcloud/vertex.glsl?
Bad reader to path: shaders/pointcloud/fragment.glsl?
Invalid shaders supplied?
Bad reader to path: shaders/splats/vertex.glsl?
Bad reader to path: shaders/splats/fragment.glsl?
Invalid shaders supplied?
Bad PLY reader: data.ply?

42yeah commented

Hi @jaccen , you should be running gsrast or any other program in the build directory (so cd ./build && ./apps/gsrast/gsrast or your build dir as cwd). Also you need to provide the trained data.ply yourself; if you don't have one yet, I have trained the train scene and you can download it here.

jaccen commented

thanks，get work。
only two model ：points and eliipsoids。 how can i render the normal pic use this project？

42yeah commented

thanks，get work。 only two model ：points and eliipsoids。 how can i render the normal pic use this project？

True Gaussian rasterization is CUDA-only and you can find the implementation in the experimental-1 branch. Related CUDA render engine source code is in apps/gsrast/gscuda.

jaccen commented
  • no plans to implement a 3d gaussian demo that doesn't require CUDA?
42yeah commented
  • no plans to implement a 3d gaussian demo that doesn't require CUDA?

Sorry, not for now. There's no good non-CUDA solution for per-frame depth-based radix sort yet. If there were, maybe I'll try to implement it.

42yeah commented

Wow, OK. I did not know that. Thanks for letting me know! Though, from what I've gathered in its README, the splat sorting is on the CPU? I'll look into it; I was planning to implement a CUDA-less version from the very beginning (but that's going to take a while).

jaccen commented

yes，i also want to implement a no -cuda with opengl c++ version，i have implented a version from this，https://github.com/mkkellogg/GaussianSplats3D, but i think there will be some optimizations should to do，like computer
cov3d in shader . maybe i will set the source on github later.

  • 👍1