stray points in the trained point cloud.

[nisarg-ujjainkar]

trafficstars

I am training the model for lego scene from nerf-sythetic dataset. Here is ow the result looks like.

Render

GT

My setup:

Code version: most recent comment. Cuda: 12.2 Pytorch: 2.5.1

I have docker setup with an older commit , which does not have this issue. How do I fix this?