search

autonomousvision / kitti360Scripts

This repository contains utility scripts for the KITTI-360 dataset.
MIT License
386 stars 63 forks source link

Inquiry about SICK laser intensity #125

Open Shanci-Li opened 2 months ago

Shanci-Li commented 2 months ago

Hi Yiyi! Thanks for your great work and effort in maintaining the repo. I am currently trying to generate the accumulated point clouds with intensity from raw 3D data. For Velodyne, I know the intensity is saved in the 4th column in the binary file. But it seems there are only 2 columns for SICK scans. Does that mean there is no intensity provided with SICK scans?

Best regards, Shanci

yiyiliao commented 2 months ago

Hi Shanci, thank you for your interest! Yes, our provided data do not include the intensity of SICK scans.

Shanci-Li commented 2 months ago

Hi Yiyi! Thanks for your help! Besides the SICK intensity, I have a question regarding an explanation about Velodyne intensity you had in another issue #73 : 'Also, values above 100 indicate retro-reflective targets.'. I checked the distribution of the intensity in the accumulated point cloud I had recovered, the normalized intensity mostly ranges from 0-0.4 (which is 0-100 for an 8-bit scale). To generalize the dataset with the high-accuracy lidar scans (intensity in 16 bits), do you think it is a good idea to cut off the intensity above 0.4 and normalize the intensity again? (I observed that the intensity from high-accuracy LiDAR generally follows a shifted Gaussian Distribution between 0-65535). I am not sure whether retro-reflective targets are important in semantic segmentation applications. I would apprentice it if you can give some suggestions on the data processing.

Best regards, Shanci