About the type of LiDAR point clouds #22
Description
Hello,
I am currently trying to apply DMD3C to LiDAR data from a SLAM system (ConSLAM), but I realized that my LiDAR is a line-scan / solid-state LiDAR rather than a spinning mechanical LiDAR like Velodyne.
In other words, my point clouds are much sparser and have a strong line-structured sampling pattern, not the dense 360° scans typical in KITTI or NuScenes.
From the paper and examples, it seems that DMD3C is demonstrated using dense LiDAR point clouds. I would like to ask:
Is DMD3C designed specifically for dense rotating LiDAR (e.g., Velodyne-style), or can it also work with line-scan / solid-state LiDAR point clouds?
Does DMD3C assume any particular LiDAR scanning pattern or angular coverage when projecting LiDAR into the image plane?
In practice, how sensitive is the method to the sparsity and structured missing regions of projected depth maps caused by line-scan LiDAR?
I have already converted my data into a KITTI-like format (RGB, depth, and calibration), but I am concerned that the domain gap between dense and line-scan LiDAR may significantly affect the performance.
Thank you very much for your great work!