How to choose raydist_fn

google-research / multinerf

A Code Release for Mip-NeRF 360, Ref-NeRF, and RawNeRF

Apache License 2.0

3.58k stars 339 forks source link

How to choose raydist_fn #25

Closed CSU-NXY closed 1 year ago

CSU-NXY commented 1 year ago

Hi, I'm using my own dataset in which near=1e-4 and far=5. I noticed that when using jnp.reciprocal as Model.raydist_fn, almost all sample points are mapped to a small range around 0.

If I set near=2 and far=6, the mapping result seems much reasonable.

Since there are five mapping functions provided in construct_ray_warps, how can I choose the most appropriate one for my dataset?

jonbarron commented 1 year ago

I think this is something of an open research question. I'd just try a few and see which works best.

bmild commented 1 year ago

I would recommend "piecewise" in the case where you need to set the near value to 0 or something very small.

CSU-NXY commented 1 year ago

@bmild @jonbarron Thanks for quick reply. piecewise works perfect for me! The mapping results seems quite good now.