Error 3D visualization

[simul-ci]

When launching CARLA, the map appears very glossy with a sky background and very low apparent resolution of buildings/pedestrians. Logs show an OpenDRIVE parsing failure and missing map, which prevents traffic lights and proper assets from spawning (not sure if it is related).

4.26.2-0+++UE4+Release-4.26 522 0
Disabling core dumps.
ERROR: unable to parse the OpenDRIVE XML string
WARNING: Could not generate traffic lights: missing map.

Steps to reproduce:

./CarlaUE4.sh -quality-level=Epic -RenderOffScreen  -nosound
4.26.2-0+++UE4+Release-4.26 522 0
Disabling core dumps.
ERROR: unable to parse the OpenDRIVE XML string 
WARNING: Coud not generate traffic lights: missing map

I think it is missing some map files (such as .xodr), which causes the very low resolution.

[InhwanBae]

Hi @simul-ci,

Thanks for your interest in our work! This is an issue I'm seeing for the first time. I conducted my experiments on a system with an RTX 4090 GPU running Ubuntu 22.04 LTS. Could you please share some information about your environment? I haven't tried running it on macOS, but I did get a Carla renderer error on a Windows system.

[simul-ci]

Thank you!
I used EC2, with an A10G GPU with an Ubuntu instance (22.04).
It uses VULKAN for rendering and I used the offscreen option.

When lauching, did you also have these errors/warnings on the server side?
ERROR: unable to parse the OpenDRIVE XML string
WARNING: Coud not generate traffic lights: missing map

[InhwanBae]

Hi @simul-ci ,

First of all, I have not used the offscreen option. Nevertheless, I think the graphic quality issue and the error/warning are separate problems. This is because although there is a dummy traffic light in the scene, it is not used at all. Rather, I think the LOD data has not yet loaded. Could you please try connecting to the EC2 instance after a sufficient amount of time has passed, or check the CPU and system memory usage?

Additionally, I am currently trying to reproduce the issue you reported. I will test it until Sunday and get back to you with an answer.
Thank you again for your interest in our work!

[InhwanBae]

Hi @simul-ci,

Thanks for waiting. I tried 3D visualization from scratch, and I found some things worth sharing.

1. After extracting Carla 0.9.15 and the CrowdES map extension and running Carla, no errors or warnings occur other than Disabling core dumps., which is intended by Carla. The errors and warnings you showed seem to have happened before loading the CrowdES map, so I think it might be a compatibility issue between Carla and the EC2 system. Could you check whether the same issue happens in vanilla Carla before applying the map extension? Since Carla itself doesn’t require a very powerful GPU, I recommend testing it on another local Ubuntu system.

2. If using the -quality-level=Epic option still doesn’t give you sufficient LOD quality, try opening the console using the ` key or F10 key and entering the following command. This forces all objects to use the maximum level of detail:

r.StaticMeshLODDistanceScale 0
r.SkeletalMeshLODBias -10
foliage.ForceLOD 0

3. Due to Carla’s internal settings, the trajectory path might have a glowing effect similar to a laser. This happens because of Carla’s postprocess. You can fix it by entering the following commands:

r.BloomQuality 0
showflag.Emissive 0

Alternatively, you can set the following colors in run_trajsim.py to 1/100 of their current values to resolve it:

Crowd-Behavior-Generation/3D Visualization Toolkit/run_trajsim.py

Lines 53 to 56 in 299ea7b

	walker_color_code = [(234, 56,41 ), (246, 133, 17), (248, 217, 4), (170, 216, 22), (39, 187, 54), (0, 143, 93), (15, 181, 174), (51, 197, 232), (56, 146, 243), (104, 109, 244), (137, 61, 231), (224, 85, 226), (222, 61, 130)]
	#walker_color_code = [(234, 56,41 )]
	vehicle_color_code = [(234, 56,41 ), (246, 133, 17), (248, 217, 4), (170, 216, 22), (39, 187, 54), (0, 143, 93), (15, 181, 174), (51, 197, 232), (56, 146, 243), (104, 109, 244), (137, 61, 231), (224, 85, 226), (222, 61, 130)]
	#vehicle_color_code = [(56, 146, 243)]

I tested in multiple ways, but I couldn’t reproduce the issue, so I’m sorry I can’t give you a complete fix. Could you check whether the same issue happens in vanilla Carla before applying the map extension?

[InhwanBae]

Hi @simul-ci,

I was finally able to reproduce the same console and visualization output as yours. In conclusion, you can safely ignore that warning, because my map file does not use an OpenDRIVE XML file and has no traffic lights. Also, I was able to get a result very similar to your screenshot. It takes a considerable amount of time for all LODs (Levels of Detail) to load after launching CARLA, and this process is heavily dependent on CPU and memory specifications.

Next, the LODs also change depending on CARLA's screen size and resolution, so forcing them to the maximum level could be helpful to solve your off-screen environment. In an environment with a screen, you can use the command I mentioned in the last comment. However, in a headless situation like yours, please force the starting map to CrowdES_hyang and then launch CARLA with the following command:

./CarlaUE4.sh -execscmds="r.StaticMeshLODDistanceScale 0, r.SkeletalMeshLODBias -10, foliage.ForceLOD 0, r.BloomQuality 0"

Finally, while troubleshooting your case, I have updated the 3D_Visualization_Toolkit/run_trajsim.py and 3D_Visualization_Toolkit/README.md files in the 3D Visualization Toolkit. This version should work much better.

I hope this resolves your issue!

[simul-ci]

Thank you a lot for your help. I was able to fix the glowing issue thanks to your help.
It seems that
./CarlaUE4.sh -execscmds="r.StaticMeshLODDistanceScale 0, r.SkeletalMeshLODBias -10, foliage.ForceLOD 0, r.BloomQuality 0"
Did not work for me (Carla 0.9.15) but I could change the settings directly by typing "`".

Regarding the quality, it has improved a little bit, but I suspect that it is caused by the GPU that is not powerful enough (A10G). I will try to run the code with a better GPU.

[InhwanBae]

Hi @simul-ci,

Thanks for the update! I'm glad to hear that the glowing issue is fixed.

Additionally, you can now fix the line colors using our new run_trajsim.py. I recommend updating your visualization code to incorporate this.

I also hope the rendering quality issue with the clay models can be resolved soon. An interesting thing I noticed in the latest screenshot (Figure 2) is that some of the trees I copy-pasted are rendering correctly, while others are not. Since they all share the same mesh and texture, the data should be available. This makes me suspect there might be a different issue, perhaps related to VRAM or another resource limit.

Figure 1: Trajectory colors fixed. / Figure 2: The copy-pasted trees.

Please feel free to reach out if you have any questions or run into any problems! Also, if you happen to find a solution for the quality issue, it would be great if you could share it for the benefit of other users.

And again, thank you so much for your interest in my work!