Inference on own LiDAR data

[JorritWillaert]

Inference on own LiDAR data

[pxiangwu]

Hi, for your own dataset, I think you can do the following two steps: (1) for a given frame at current timestamp, use the calibration information to align the past frames with the current one. (2) preprocess the point clouds into BEV maps, and feed these BEV maps into the network.

I think even if your LiDAR data is not 360°, the network can still predict as expected. This is because the final output will be masked by the occupancy information. That said, if a given grid cell is not occupied by some LiDAR points, the prediction at this cell will be thresholded to zero (i.e., background).

Similarly, it does not matter if the angle is tilted, as long as such a tilted angle is not very large. All you need to do is provide the BEV maps into the network.

[JorritWillaert]

Hi,

Thank you very much for your reply! I will definitely try this out in the upcoming days.