search

grasp-lyrl / NeRFtoGSandBack

Apache License 2.0
46 stars 1 forks source link

Exporting Generated Scene #6

Open karthik101200 opened 3 months ago

karthik101200 commented 3 months ago

Hello. Thanks for the code. I tried nerfgs for a custom dataset and the results are pretty good. I want to export the scene to a mesh or any other dense representation which can be later imported in a simulation environment. Wanted to ask if there is a way to do so. Thanks.

Also just to add, I didnt run nerfsh first as give me out of memory error ns-train nerfsh --data DATA-PATH --pipeline.model.camera-optimizer.mode off

RuntimeError: /tmp/pip-req-build-sf4k0pds/include/tiny-cuda-nn/gpu_memory.h:584 cuMemCreate(&m_handles.back(), n_bytes_to_allocate, &prop, 0) failed: CUDA_ERROR_OUT_OF_MEMORY

Any parameters to tune so that I can train. I have RTX A2000 GPU

boooolean commented 2 months ago

For exporting the nerfgs, it is currently only directly exportable as a .ply file similar to splatfacto. There are tools you can find online to convert the .ply file into dense representations.

As for nerfsh running out of memory, I am not sure, we will have to look more into it, but similar parameter tunings to standard nerf will also help out. Out of curiosity, how did you run nerfgs without running nerfsh? In our proposed pipeline, we had assumed you first create a nerfsh and then convert it into a nerfgs.