Why only top lidar?

[fcakyon]

I have to say that, this repo is really useful for using the waymo dataset on non-tf frameworks :) However when i look for the visualise_labels_and_lidar, groundtruth_extraction scripts I see they only use top lidar points instead of combining all lidar point clouds. Am I missing something or is it really the case?

[gdlg]

Thanks, I am glad that it’s useful.

I have never tried to combine the information of multiple LIDARs, though there is no reason it wouldn’t work. I was using the LIDAR to check which object is occluded or not.

It shouldn’t be too hard to loop through all the LIDARs and combine the information. I would be happy to accept a pull request if you have a look at it.

[fcakyon]

Thanks for the reply, I will consider doing an pr.

Another question: At this line, by saying "box space" did you mean "global frame"? I'm taking the coordinate system given in Coordinate Systems section as reference.

[gdlg]

No, this is a coordinate system set at the centre of a given label’s bounding box and x-axis pointing toward the front of the vehicle, y-axis toward the side and z-axis up. Each axis is scaled such that in this frame, a point is inside the bounding box if and only if all its coordinates are in the range [-1,1].

This is used at this line to determine which LIDAR point falls within the bounding box of each label and count the number of LIDAR points per label. Alas this involves a cartesian product here between each LIDAR point and each label. That’s why this code is quite slow.