SJoJoK / 3DGStream

[CVPR 2024 Highlight] Official repository for the paper "3DGStream: On-the-fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos".
https://sjojok.github.io/3dgstream
MIT License
302 stars 18 forks source link

Clarity on instructions #9

Closed HarshitGupta29 closed 4 months ago

HarshitGupta29 commented 4 months ago

Follow the instructions in gaussian-splatting to get a high-quality init_3dgs (sh_degree = 1) from the above colmap dataset, which will end-up like:

Final init_3dgs gaussian point cloud has sh_degree =3. Is this a typo or we should only train for 1k iterations to keep sh_degree equal to 1 ?

SJoJoK commented 4 months ago

This means that the argument --sh_degree should be set to 1. We will clarify it in the README, sorry for any inconvenience.

HarshitGupta29 commented 4 months ago

Thanks a lot!