Open hturki opened 2 years ago
I left them constant (1/num_of_pts_on_a_ray) in my training.
BTW, you should use the normalized distance instead of the Euclidean distance (see Eq.11 in mip-nerf 360 paper).
I left them constant (1/num_of_pts_on_a_ray) in my training.
BTW, you should use the normalized distance instead of the Euclidean distance (see Eq.11 in mip-nerf 360 paper).
could I just leave it as zero? many thxs
Hi,
What do you set the interval value to for the last sampled point on each ray. Using a large term (1e10) as a substitute for infinity make the loss term very large.