How can we generate the dense depth image from sparse LiDAR points?

[ZhenboSong]

Hi Hengli,
Could you provide some instructions for generating dense depth maps? Using depth completion networks or simply using interpolation? If we don't have perfect depth maps, could SNE and RoadSeg Net generalize well?

[hlwang1124]

Hi Zhenbo. In our experiments, we simply use interpolation.
In addition, we have not tested our approach under different noise levels. However, I think our approach can generalize well on the commonly used sensors, such as LiDAR for self-driving cars and RealSense for mobile robots.

[ZhenboSong]

Thanks, Hengli. Could you provide the code for interpolation? We have tried several depth completion methods, but RoadSeg results are not good as you provided in your videos.

[hlwang1124]

Hi, Zhenbo. Actually, the code we use is very simple, griddata method in Matlab, which is also equivalent to scipy.interpolate.griddata in python. We have provided the dense depth maps in the KITTI Road Dataset, and you can also have a look.

[ZhenboSong]

Great! Thanks again.

[cattpku]

Hi @ZhenboSong, have you successfully generated the uint16 depth image from lidar data? If so, can you kindly share the script? Thanks.