Ruilong Li(李瑞龙)

This is weird. May I know your CUDA and pytorch version? ```bash python -c "import torch; print(torch.__version__)" nvcc --version ``` And what's your GPU? ```bash nvidia-smi ``` My guess this...

Sorry I mean which NVIDIA card are you using, e.g. V100?

I can reproduce this error with torch 2.0.0. With torch 1.13.0 everything seems working fine. I'll come back to this issue once I figure out why this happens. In the...

Just fixed it on the master branch!

what is the parameters you used for ray_marching? It might be the case you are creating too many samples, which causes overflow. Currently the number of samples in each batch...

Thanks @DreekFire for the updates. This is not an easy one. My main question is on the fisheye formula. Have you run any tests to verify the fisheye formula is...

Profiling using [line_profiler](https://github.com/pyutils/line_profiler) with commands: ```bash CUDA_LAUNCH_BLOCKING=1 CUDA_VISIBLE_DEVICES=5 kernprof -l scripts/train.py nerfacto --data data/nerfstudio/poster --max-num-iterations 1000 --viewer.quit-on-train-completion True ``` Locates `nerfstudio.cameras.radial_and_tangential_undistort` to be the current bottleneck. ```bash Timer unit: 1e-06...

Nerfacc takes care of ray undistortion. data loading consumes 43% -> 34% ```bash Timer unit: 1e-06 s Total time: 5.89843 s File: /home/ruilongli/workspace/nerfstudio/nerfstudio/cameras/cameras.py Function: generate_rays at line 312 Line #...

PR Summary: 1/ Update Instant-NGP (tested) and Nerfplayer-NGP (untested) model with much better performance: - Switch from `TCNNInstantNGPField` to `TCNNNerfactoField`, which is actively maintained. In this commit the `TCNNInstantNGPField` is...

Not sure what to go with further speedup, because: - Precompute all rays would be beneficial only if there is long enough training (so that every ray is trained more...