About data generation

[autonomousvision]

About data generation

[ap229997]

The evaluation statistics indicate that the agent has completed the route which means that the data generation is working fine. The data is stored at the location assigned to SAVE_PATH in leaderboard/scripts/run_evaluation.sh.

[XiangTodayEatsWhat]

Thanks for your quick reply! According to your reply, I have found the generated images. They look different from the downloaded data in structure, but it probably doesn't matter. Thank you again sincerely!

[ap229997]

Can you tell the difference in structure in the downloaded and generated data? Is it just the directory names or anything else too?

[elegentbamboo]

Can you tell the difference in structure in the downloaded and generated data? Is it just the directory names or anything else too?
My generated data has only one extra semantic_lidar directory compared to the structure of 14_weathers_data as you described.

[ap229997]

That's fine. I later added the functionality of semantic lidar in the data generation script.

[XiangTodayEatsWhat]

Sorry for the late reply, you are right. It is just the difference in directory names.

[ZKL-VOILET]

Hello! I have some problems about the data generation.

I have no idea about this problem. Could you please help me with the problem?
thanks

[ap229997]

From the error, it seems like the CARLA server cannot bind to port 2000 since it is already in use. Can you verify if the port is available or not?

[ZKL-VOILET]

From the error, it seems like the CARLA server cannot bind to port 2000 since it is already in use. Can you verify if the port is available or not?

Thanks for your reply. I have changed the port to 8000, but it stuck here for half an hour. Dose it work?

[ap229997]

This message is normal, the server should be running now. You can try connecting a Carla client to the server and verify.

[Kin-Zhang]

Hi, @ap229997 I found that about the data generation agent, did you find it not so correct also about the npc_agent:
I tried this scenario and find things interesting:

Like here the auto_agent and npc_agent will have collision and I notice there are many collision test fail on Town03, did you find that maybe a reason for the agent cannot avoid this kind of case?

It just my personal find out, since I tried to do some things more based on this repo.

[ap229997]

I did not experiment with the npc_agent so I don't know about its failure cases.

Regarding our autopilot, I am aware that it collides a few times. This is because our autopilot uses several heuristics for collision avoidance which has certain blindspots.

[Kin-Zhang]

I did not experiment with the npc_agent so I don't know about its failure cases.

Regarding our autopilot, I am aware that it collides a few times. This is because our autopilot uses several heuristics for collision avoidance which has certain blindspots.

Thanks for the quick reply! I see.
Thanks again.