Hi. I don't know how many people actually go crazy with crowds and render them but after simulation fun, all the flipbooks, there's rendering nightmare. I solved many issues but this just beats me - does anybody know what could be the cause of random parts of the agents disappearing? The smaller ones, knights, are taken from Mixamo and they never flicker, but the bigger ones, baked before and and all that stuff, just go crazy. Some of them are ok the whole time, some of them partially disappear and reappear. I'm out of ideas. They're ok in the viewport.

Some short videos:



I've attached a screenshot.

Is it the scale problem? (they have a scale of 3, with some randomization). But why do we have scale then… I put them in separate network, so after the sim each group is in its own GEO node, because at first it seemed to solve the problem, but no - different frame and it's back again. I've reopened the scene, restarted the computer, rendered several times. I've removed the shader, changed it to simple Principled Shader without any adjustments - nothing helps

Any tips are welcome.

My guess would be that there are bad bounding boxes for the agents - does Mantra give you any warning during the render about out of bounds geometry?

Ok, I think that's a good guess Thank you! I put Verbose level to 3 and see the messages:

Procedural generated out of bounds geometry
Actual Box: (-14.1973, -24.3555, 24.4797) - (-10.0625, -17.6432, 30.9894)
Specified Box: (-13.5555, -24.2203, 24.9912) - (-10.8644, -21.4461, 27.9729)

This message is also printed for the smaller agents, which render properly - why is that? I would never guess that New area to me. Especially that the same agents render in different ways - sometimes only the legs, sometimes the torso with the legs. It varies. Shouldn't the bounding box of a given agent always be the same?

If you could elaborate a bit cwhite I would really appreciate that

Edit: Ok, I've read some more here:
http://www.sidefx.com/docs/houdini/crowds/agents [www.sidefx.com]
In the section about the bounding box it says it will use a collision layer if I have one. I started building one for that agent - but I added collision shapes only to the legs - so I assume this is why the bounding box is usually only in the lower parts, right? But sometimes the whole agent is visible… Shouldn't that be more consistent?

Well - how can I force the bounding box to take the whole agent into consideration. Also the smaller agents give errors - and they have they're whole body set up for collisions… Is it because of the objects attached to their hands?

How can I get rid of the errors?

Can I change agent current layer or influence the bounding box post-sim?

Thank you so much for any tips.

You can change what layer is used as the current layer or collision layer at any point. The Agent Edit SOP is one way to do that.

I've tried it before you posted but didn't succeed. However, now I did it again setting both current layer and collison layer to “default” (I don't need the collision layer anymore, because the sim is done) and on one of the frames it works. I'll set up an overnight render to see how it goes.

Thank you for your swift reply. If there are any other ways, maybe more “proper” ones, I would love to learn about them.

EDIT: It helped a lot but there are some frames that still have clipping geo. I couldn't continue - no more time. I'll try to get back to the project in the future. That's often a problem for me with Houdini - some small things that stop me from using it in fast paced production - usually those things are not mentioned in official videos or the manual (this clipping problem should be stressed somewhere).
Here's another possible solution:
http://forums.odforce.net/topic/32119-crowd-sim-missing-parts-in-render-random-and-changing-between-frames/ [forums.odforce.net]

Yeah, Ive got a similar problem coming from the Agent Collision Layer SOP. It clips my agent during rendertime. I just cant figure out a way to get around this. Only way is to disable the collision layer but then I cannot use ragdolls in my sim. Ridicilous. HELP

You can change which layer is used for the agent's collision layer at any point (e.g. post-sim) with the Agent Edit SOP. The collision layer doesn't need to be named ‘collision’

Hi all!
That collision layer made my life hard but I realized that you can add a collision layer after simulation just for this issue. I created a boundingbox cube layer and added it after simulation.

Just remember to untick the “Current Layer” and “Collision Layer” in the agent setup step so it wont mess with ragdoll simulation or viewport display.