Question about Carla Settings for simulate a specific lidar

[DavideCoppola97]

Hi, in the paper you simulate a VLP16 lidar using the Carla simulator and I have a question about that.
I tried to simulate a Velodyne HDL64E with Carla and I got the related pointclouds, aesthetically they look similar to those obtained with real lidar (like those of SemanticKitty dataset) but when I go to make inference on these pointclouds with some of the networks for semantic segmentation most famous (SqueezeSSegV3, SalsaNext, 3DminiNet etc) I get almost completely wrong labels, the networks cannot correctly recognize even the road. I doubted that I am doing something wrong in simulating the lidar in Carla, although I have entered, I believe, the same parameters of the lidar in question:
Here are the parameters I use to create my pointclouds with Carla (executed asynchronously with parameter -fps = 10):
lidar.set_position (0, 0.0, 1.73) # 1.73 meters should be the height from the ground of the lidar used in SemanticKitty, however I have the same results even with height = 0
lidar.set_rotation (0, 0, 0)
lidar.set (
Channels = 64,
Range = 120,
PointsPerSecond = 1250000,
RotationFrequency = 10,
UpperFovLimit = 2, #tested also with 3
LowerFovLimit = -24.9) #tested high with -25

So I would like to know how you have simulated your lidar, maybe there is some parameter that escapes me, I also leave attached the datasheet from where I got the parameters. What I think may be wrong is the range or ratio between the server's fps and the RotationFrequency parameter.