search

mileyan / Pseudo_Lidar_V2

(ICLR) Pseudo-LiDAR++: Accurate Depth for 3D Object Detection in Autonomous Driving
MIT License
603 stars 126 forks source link

How to generate point clouds like the one in the paper #28

Open rsj007 opened 2 years ago

rsj007 commented 2 years ago

I have generated point clouds using "generate_lidar_from_depth.py" with depth maps. I visulized the point clouds by mayavi, however very different from the one in the paper. The point clouds I generated has red distorations (inside the red circle)and some long tails. Anyone knows how to get the point clouds properly? 1145df43e212b73cf0f561e27fc23bf

image

Baboom-l commented 2 years ago

This part may be caused by a wrong depth estimate in the sky. You can try to set the maximum height to 0.2m in generate_lidar_from_depth.py.

rsj007 commented 2 years ago

Thanks! Is it 0.2m too small for visualize cars? And, which visualization tools do you use? Can you get ideal point clouds output?

Baboom-l commented 2 years ago

For the sedan, it does not affect the visualization results. However, it may have an effect for trucks or off-road vehicles in close proximity.These suggestions are only for the visualization of the point cloud and I am not sure if they will affect the results of the subsequent 3D object detection.

Baboom-l commented 2 years ago

I use the visualization in OpenPCDet