dcharatan / pixelsplat

[CVPR 2024 Oral, Best Paper Runner-Up] Code for "pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction" by David Charatan, Sizhe Lester Li, Andrea Tagliasacchi, and Vincent Sitzmann
http://davidcharatan.com/pixelsplat/
MIT License
830 stars 56 forks source link

How to generate point cloud in world coordinate? #55

Closed railgun526 closed 5 months ago

railgun526 commented 6 months ago

Hi @dcharatan, thank you for helping me solving so many questions! I have a small question to ask. I want to generate point cloud in world coordinate in 3D space and use open3d to visualize it. I found that you have written generate_point_cloud_figure.py but it generate point cloud which is projected to 2D plane. I am wondering if the means of the gaussian can represent the point cloud? By simply doing so, I got the result which may not seem satisfactory... 微信图片_20240307001440 Could you please give me some clues about how to convert gaussians into point cloud in world coordinate in 3D space? Thanks a lot!

dcharatan commented 5 months ago

See #43 -- in practice, it can be helpful to filter out far-away Gaussians, since the model tends to create a lot of junk points behind the correct, 3D-consistent scene.