search

ashawkey / torch-ngp

A pytorch CUDA extension implementation of instant-ngp (sdf and nerf), with a GUI.
MIT License
2.11k stars 275 forks source link

num_rays(N) is 2073600, num_sigmas is 2073728? #170

Open BCL123456-BAL opened 1 year ago

BCL123456-BAL commented 1 year ago

1 run ''python main_nerf data/nerf_synthetic/chair --workspace chair -O --bound 1.0 --scale 0.8 --dt_gamma 0 --gui''

  1. code in nerf/renderer.py
  2. questions: N = rays_o.shape[0] //N=2073600 ... while step < max_steps: //step==1 xyzs, dirs, deltas = raymarching.march_rays(...) //xyzs==Tensor(2073728,3) sigmas, rgbs= self(xyzs, dirs) //rgbs==Tensor(2073728,3), sigmas ==Tensor(2073728,1)
  3. Why the number of generated sigmas will be 128 more than the number of rays？