NVIDIAGameWorks
Running out of memory when using kaolin functions
Hello, I am using kaolin to transfer triangle meshes to voxel grids and then perform operations on them. The issue that I am having is that when I convert the voxels back to triangle meshes and try to call operations such as uniform_laplacian_smoothing it tries to allocate significantly more memory than is available to it and fails, this happens both on CPU and CUDA. On top of this, if the meshes are small enough that it does performs the operations, it does not de-allocate the memory and I need to restart the environment to do so. My question is, is there anything that can be done to limit the memory these functions allocate or will I need to refrain from using certain features because of the complex meshes that I am voxilizing
Thank you!
appying marching cube on a high resolution voxelgrid will indeed generate a lot of triangles and might be difficult to compute laplacian matrix (at least dense, sparse might be ok).
I'm not sure about the memory not being de-allocated, pytorch usually has its own garbage collector, do you have a specific tensor in mind?
What resolution of voxelgrid are you using? Also what GPU do you have?
