fabiotosi92 / NeRF-Supervised-Deep-Stereo

A novel paradigm for collecting and generating stereo training data using neural rendering
https://nerfstereo.github.io/
MIT License
349 stars 19 forks source link

[About 3DGS Supervised Deep Stereo] #47

Closed CaptainEven closed 8 months ago

CaptainEven commented 8 months ago

Hi, 3DGS has been prevalent nowadays, May i ask any palnnning to replace the NeRFs with 3DGS?

fabiotosi92 commented 8 months ago

Hi!

We have explored replacing NeRFs with 3D Gaussian Splatting but found that, despite 3DGS has faster inference times compared to NeRF variants, and sometimes better rendering quality, it still suffers from similar (or sometimes worse) geometric structure limitations compared to NeRF-based solutions, so depth/disparity supervision using 3DGS does not provide significant advantages at the moment. We will continue to monitor 3DGS developments for potential future incorporation.

CaptainEven commented 8 months ago

Hi!

We have explored replacing NeRFs with 3D Gaussian Splatting but found that, despite 3DGS has higher inference times compared to NeRF variants, and sometimes better rendering quality, it still suffers from similar (or sometimes worse) geometric structure limitations compared to NeRF-based solutions, so depth/disparity supervision using 3DGS does not provide significant advantages at the moment. We will continue to monitor 3DGS developments for potential future incorporation.

Thanks for the exploration and explanation!