New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Improve Gaussian Splats performance #6344
Comments
We're doing splat sorting on a separate thread, and that bit is pretty fast. - faster than 4fps for sure I think. 1M splats in about 4ms on M1 Max. |
The sorting is definitely the slowest part in my experience, it causes a lot of hitching as the gpu upload locks rendering so the fps meter might be reporting a good fps but its dropping frames. |
related: #6352 |
Recently Adobe researchers published a paper called GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting.
The page contains real-time demos featuring this viewer: antimatter15 / splat.
What's interesting about it is that it produces solid 60fps on M1 Air even with advanced scenes, while the same PLY imported into PlayCanvas results in less FPS (with Device Pixel Ratio unchecked).
For example:
https://sai-bi.github.io/project/gs-lrm/viewer/index.html?is_object=false&url=https://huggingface.co/datasets/saibi/gs-lrm/resolve/main/scene_assets/002000_gaussians.ply
https://playcanv.as/index/7d0ab8f5 (https://playcanvas.com/project/1214533/)
From viewer description it looks like it's doing splat sorting on CPU in a separate worker thread on async basis:
I guess PC might introduce similar configuration for 3DGS, with selector in inspector allowing to choose a sorting algo.
The text was updated successfully, but these errors were encountered: