search

autonomousvision / mip-splatting

[CVPR'24 Best Student Paper] Mip-Splatting: Alias-free 3D Gaussian Splatting
https://niujinshuchong.github.io/mip-splatting/
Other
1.01k stars 65 forks source link

How to test 3D-GS+EWA？ #30

Closed kaka-1314 closed 5 months ago

kaka-1314 commented 5 months ago

Thanks for your great job!Due to both 3D-GS+EWA and 2D mip filter have the same formula, I wonder how can we change the code to get the code of 3D-GS+EWA? Just delete the code of 3D filter can realize the target?

I am looking for your reply, thank you very much! image

kaka-1314 commented 5 months ago

Thanks for your great job!Due to both 3D-GS+EWA and 2D mip filter have the same formula, I wonder how can we change the code to get the code of 3D-GS+EWA? Just delete the code of 3D filter can realize the target?

I am looking for your reply, thank you very much! image

In my opinion, the difference between EWA and 2D mip is the number of kernel_size, is it correct? If my thought is correct, the difference between 3D-GS+EWA and mip-splatting w/o 3D filter is also the number of kernel_size, is it?

niujinshuchong commented 5 months ago

Hi, Yes. You can remove the 3D smoothing filter by setting it to zero here https://github.com/autonomousvision/mip-splatting/blob/main/scene/gaussian_model.py#L190 and use the kernel size of 0.3.