princeton-computational-imaging / SeeingThroughFog

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

What would be the reason of missing detection in front of velodyne 32c?

happengor opened this issue · comments

Hello STF team,

First of all, thank you for your contribution, this is a amazing dataset. But I have a question, I'm wondering why there is missing detection in front of the vlp32c in many frames, I'm expecting unwanted detections of snowflakes and road reflection.

Screenshot from 2022-11-02 10-35-08

Best regards

Hi @happengor,

thank you for your kind words and interest in our work.
The VLP is not externally rotating and has an enclosed plastic cover. Therefore, it is not able to "spin-off" dirt as wet snow particles building up infront of the sensor. In contrast the none enclosed HDL spins off such dirt. Active wiping is prohibitiv as they might scratch the plastic housing leading to undesired laser deflections. As such the SOTA is to keep them clean through the aerodynamics of the car. On roads as shown in the picture or in case of narrowly overtaking cars the sensor is clogged up very soon. We cleaned the sensor between recordings, but as this build up is so quick we decided to keep it in the dataset as it represents the current SOTA.

BTW Figure 13 in this paper shows the LiDAR in those conditions