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

Multi resolution training #24

Closed KeKsBoTer closed 5 months ago

KeKsBoTer commented 5 months ago

Hey,

Thanks for publishing your code! In the paper, you write that you train and test on multi-resolution images (40% full resolution). In the code this is activated with the "sample_more_highres" flag. But this flag is only set to True for the synthetic data. Am I missing something or do you not sample different resolutions for the other datasets?

Thank you!

niujinshuchong commented 5 months ago

Hi, Yes, we have multi-resolution training only on the NeRF-synthetic dataset. For other experiments on the paper (single-scale training and testing on mip-nerf 360 dataset and single-scale training multi-scale testing on both NeRF-synthetic and mip-nerf 360 dataset), we only use a single scale images, so it is not used.

KeKsBoTer commented 5 months ago

Thank you for the quick reply!