search

unmannedlab / RELLIS-3D

RELLIS-3D: A Multi-modal Dataset for Off-Road Robotics
https://unmannedlab.github.io/research/RELLIS-3D
265 stars 47 forks source link

Intensity Normalization #36

Open seungtae-Han opened 9 months ago

seungtae-Han commented 9 months ago

Hello RELLIS-3D,

I am currently training a deep learning network for point cloud segmentation using Rellis-3d, and I am conducting inference on point clouds that I have obtained using my lidar sensor. The lidar I use (both Rellis-3d and the lidar I use) is os1 64.

The lidar point cloud I have acquired has intensity values greater than 1. However, the intensity in Rellis-3d is observed to be between 0 and less than 1, and I'm curious about how the values were normalized! (For example, did you acquire lidar data and then complete Rellis-3d by dividing all intensity values by the largest intensity value?)

This is a visualization of the intensity distribution from a randomly selected bin file in the Rellis-3d dataset: image

And this is my lidar data: image

kasiv008 commented 5 months ago

The rellis 3D intensity data is normalized from 16 bit. You would need to multiply the data with 65535 to get the raw intensity value.

seungtae-Han commented 4 months ago

The rellis 3D intensity data is normalized from 16 bit. You would need to multiply the data with 65535 to get the raw intensity value.

Thank you!!! You've been a great help to me.