search

cvlab-kaist / RAIN-GS

Code for "Relaxing Accurate Initialization Constraint for 3D Gaussian Splatting" by Jaewoo Jung, Jisang Han, Honggyu An, Jiwon Kang, Seonghoon Park, and Seungryong Kim
https://ku-cvlab.github.io/RAIN-GS
MIT License
298 stars 16 forks source link

Slv and low pass code location #7

Closed ishaybee closed 7 months ago

ishaybee commented 8 months ago

Hi In the paper you are saying that it is a 1 line modification for your method. Can you point out where are the location of these modifications?

renhaofan commented 7 months ago

@ishaybee https://github.com/KU-CVLAB/RAIN-GS/blob/4f72d61788347fe09da5edb395d8ab080aabae5e/submodules/diff_gaussian_rasterization/cuda_rasterizer/forward.cu#L99

crepejung00 commented 7 months ago

Hi, sorry for the late reply.

  1. Sparse-large-variance (SLV) random initialization can be implemented by just changing the initial number of Gaussians! The change is implemented in line 160 implicitly.

https://github.com/KU-CVLAB/RAIN-GS/blob/4f72d61788347fe09da5edb395d8ab080aabae5e/scene/dataset_readers.py#L159-L181

  1. The progressive Gaussian low-pass filter control can be found here: line 91~97 is a simple definition of the low-pass filter value which is written in multiple lines for readability. This low-pass value is then added to the covariance as @renhaofan mentioned.

https://github.com/KU-CVLAB/RAIN-GS/blob/4f72d61788347fe09da5edb395d8ab080aabae5e/train.py#L90-L102

Thanks!