Justin Kerr
Justin Kerr
what machine are you testing on? is there any particular sort of hardware this implementation is optimized for? eg would you see stronger benefits on weaker CPUs or GPUs, etc
It would be good to compare these camera grads to the camera grads from the torch implementation with autograd, [this PR](https://github.com/nerfstudio-project/gsplat/pull/118) adds a torch implementation which should have correct (although...
I'm thinking of merging #127 because of how much simpler it is and it seems to give good results, @jh-surh thoughts?
Make sure gsplat is updated to the latest version and this should go away
what CUDA compute capability is your GPU? @liruilong940607 found that the environment variable `TORCH_CUDA_ARCH_LIST` incorrectly had old compute capabilities that prevented compilation, you can try `unset TORCH_CUDA_ARCH_LIST` to see if...
Should we make this the default for rasterizing ND things?
Testing on a 4090 I get pretty much the same speed between default ND rasterization and the index-based one: `CUDA_LAUNCH_BLOCKING=1 python tests/test_rasterize.py --profile --N 1000 --D 256` produces 429 and...
Interesting, I also tried it in nerfstudio and it runs out of memory (requests like 70GB of memory or something), is there a way to get the footprint down or...
#130 works towards this issue, let me know if you try it out! @zubair-irshad
Cool! have you tried this in 3D scenes with supervision? For API I think including `render_depth` is reasonable, although it might be annoying if there's an eventual blowup of the...