search

YanjieZe / 3D-Diffusion-Policy

[RSS 2024] 3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations
https://3d-diffusion-policy.github.io
MIT License
375 stars 33 forks source link

Image to 3D point cloud? #4

Closed DrBwts closed 6 months ago

DrBwts commented 6 months ago

How are you converting your 2D camera images to 3D point clouds?

You don't go into any detail about this in your paper or provide a reference for the method. I was wondering if you could either elaborate or point me to the code that deals with this method?

YanjieZe commented 6 months ago

Depths could be directly converted into point clouds with camera parameters. It is not a method we propose or introduced by others, but one basic truth in 3D vision.

There are some online blog that introduce this, such as this.

DrBwts commented 6 months ago

I checked out your link & I'm not convinced you can calculate depth like that as equation 1 is not unique.