search

kazuto1011 / r2dm

LiDAR Data Synthesis with Denoising Diffusion Probabilistic Models (ICRA 2024)
https://kazuto1011.github.io/r2dm
MIT License
45 stars 5 forks source link

Problem upsampling lidar #10

Closed TristanCot closed 2 months ago

TristanCot commented 2 months ago

Hello, I can get it to work by using completion_demo, but I can't get what I want and I need your help.

I'm trying to upsample a point cloud that is at 32 beams. In the first case, I put the point cloud in its 64-beam version, I apply a mask that removes one beam out of 2, and we can see the completion that takes place and therefore the upsampling. image

But when I input a 32 beams point cloud directly, completion does not take place and so I have no upsampling, as can be seen: image

Do you have any idea how I could go about upsampling directly from a 32-beam point cloud?

kazuto1011 commented 2 months ago

Have you made the mask that indicates whether to fill? If not, please define the mask, such as (depth > 0).float(), and feed it into the repaint function; otherwise, the function will leave all the pixels as is. https://github.com/kazuto1011/r2dm/blob/ad66e71e86051d0341e096c60ece30a7de6c5b3f/completion_demo.py#L94

TristanCot commented 2 months ago

Have you made the mask that indicates whether to fill? If not, please define the mask, such as (depth > 0).float(), and feed it into the repaint function; otherwise, the function will leave all the pixels as is.

https://github.com/kazuto1011/r2dm/blob/ad66e71e86051d0341e096c60ece30a7de6c5b3f/completion_demo.py#L94

Thank you very much, that's what I was missing, thank you for your work!