search

kujason / ip_basic

Image Processing for Basic Depth Completion
MIT License
564 stars 87 forks source link

How does this solution perform on lower-beam lidars? #8

Closed mac137 closed 4 years ago

mac137 commented 4 years ago

Hello! As in the topic. Could anyone, please, tell me how this solution performs on lower-beam lidars, especially 16-beam Velodyne?

kujason commented 4 years ago

The performance will depend on several factors such as the sparsity after projection into the image frame, how much of the scene structure is available in the LiDAR points, and what sort of results you are looking for (e.g. evaluation metrics and/or structure preservation). As the sparsity pattern will be different for a 16 beam Velodyne, you may need to experiment with the kernel sizes or shapes to get better results. I have tested this with a 32-beam LiDAR and with some minor adjustments, it gives reasonable results.

zhouzheyuan commented 3 years ago

@kujason Hi! how is your setting for a 32-beam LiDAR? I have used the default setting but the result doesn't feel right.

I have tested this with a 32-beam LiDAR and with some minor adjustments, it gives reasonable results.

kujason commented 3 years ago

@SirJamie I can't recall the exact settings I used, but I think I simply adjusted the initial dilation kernel to be a taller diamond shape in order to account for the larger gap between the LiDAR rings. You may need to do some experimentation to find a good result with your own setup.