Crowd system development and ideas from you

[chrispettersen]

Hey there

 

I am in the middle of developing a crowd system for 3ds max and rather than just create it and release it, i thought i might ask you guys what features you would find useful in everyday production.

 

About myself:

I have been in the arch viz field for over 20 years, and every time i am asked to populate a space with people, i run into the same issues of having to find good models, rig them, animate, and populate the scene.

Most of the time the project hasn't got the budget for this and it takes way more time to do than it should.

This is why i am putting this crowd system together, to help me with my projects, and rather than keep it to myself, i am want to release it to other studios.

 

The problem:

The current offering of 3d people look really bad, stand out too much with weird clothing, are expensive if you want lots of variations, come either only t-posed or rigged with biped which crashes often.

Solutions to crowd systems via plugins that i have tried have also lead to lots of crashes and corrupt files.

 

This is my checklist:

1. Good for medium to far distance
   
2. efficient rendering of 1000's of people
   
3. intelligent avoidance behavior
   
4. walking and standing people
   
5. Random Variation
   
6. Neutral clothes so no one stands out
   
7. No plugins, so no crashes
   
8. No network rendering limitations
   
9. easily created in less than an hour
   

 

The Tech behind it:

To do this without plugins (so that it is easy to install and network render and most importantly wont crash), i am using particle flow to drive the crowd system. So far i have been able to simulate 1000's of people walking, standing in groups, avoiding each other, avoiding obstacles and leading them to areas of interests.

 

I am creating a bunch of medium res people, with separate objects for pants, shirts, skirts, hair etc that will be randomized by about 10-20 materials, so no 2 people will look alike. Animation is hand animated from filmed reference and point cached to minimize complexity and avoid scene scale issues. Animation time will be randomized over 2-3 minutes so that you can do longer shots and nobody is doing the same thing.

 

Setup time for me has been in the minutes, i am assuming it will take someone new to it under an hour to get a crowd sim working in their scene. The Particle flow is quite complex, but I will be supplying a range of scenarios that you can merge into your project and tweak very easily by adjusting no more than 2-3 attributes.

 

This crowd system is designed for medium to far distance, as i feel that 3d people wont be real enough just yet to be believable close to the camera. For close to camera people you should use other solutions such as green screen actors or high res scanned people. I think i will release it for around the 500 dollar mark.

 

I would love some feedback on what you would like to see before i finish it off in the next week or three. If you have any questions, fire away. If you have any requests, i will see what i can do.

 

 

Thanks for reading, lets discuss.

[Nic H]

that all sounds very good - if you can do it for that price that is even better.

 

the only thing im unsure about is particle flow because its so bloody slow as soon as you get a bit complex. if you have some caching technology so we can preview the paths and movements that would be great.

 

having some sitting people would be good as well for those larger urban areas.

[chrispettersen]

Yes you are correct, particle flow can get quite bogged down with a lot of people, the way i am combating this is by using a low poly representation in the viewport, this allows up to a hundred or so characters to walk around in realtime giving you a chance to see if they interact properly, when i use more than a hundred i can either switch off the animation or even just turn to box shapes, giving me realtime performance on thousands of particles.

 

Here is a video capture of the viewport performance (keep in mind this is unfinished development work as the characters are still being built and animated)

 

[Devin Johnston]

Speed would be a big issue for me as well as memory usage, I can imagine 1000's of people eating up a lot of RAM. I know you want to make the people less detailed but I'd suggest including at least a few people that can be used in close up shots. It would be nice if they could have some limited interaction with objects, so they don't walk through walls or floors and could even go down or up stairs.

[Jeff Mottle]

Curious how this might compare to Gepetto?

[chrispettersen]

Curious how this might compare to Gepetto?

 

I was very excited when i heard of Project Geppetto, but after testing it was a bit let down by the amount of crashes i got, to the point where it was almost useless on a deadline, hopefully they will have fixed things when it is done. Also the only way to simulate the crowds is to draw pathways for walking or mark areas for standing people. there was no way to populate a large space with characters randomly walking and standing around.

To be honest its Geppetto that made me want to create my own crowd sim with many of its features, but with no crashes, plugins or max version dependencies.

[chrispettersen]

Speed would be a big issue for me as well as memory usage, I can imagine 1000's of people eating up a lot of RAM. I know you want to make the people less detailed but I'd suggest including at least a few people that can be used in close up shots. It would be nice if they could have some limited interaction with objects, so they don't walk through walls or floors and could even go down or up stairs.

 

The models will come with low, med and high poly settings, so depending on how far away you are you can use the appropriate res people. for instance using the low poly versions on aerial shots will allow you to render 10's of thousands.

Its easy to have no collision and avoidance of objects, however stair climbing is very tricky to implement and i doubt we will see a solution to this for many years yet.

[chrispettersen]

 

having some sitting people would be good as well for those larger urban areas.

 

 

Good idea, will add this i think

[Nic H]

is gepetto '3dsmax design' only?

[Frosty]

Isn't this what you're talking about?

http://www.axyz-design.com/anima/

[Francisco Penaloza]

I really like the idea of a small developer for task like this, and I really don't like bubble poppers, you have excellent samples like Building Generator, or Ghost town; but I tried before use particle system to animate people, and it get's really slow! I am not a developer in any way and maybe you are using some code to optimize this??

you mentioned that you'll use point cloud for people animation? that's a lot of over head on the memory too, time ago I did a giant fish tank animated, with around 500 fishes, low poly simple tale wiggle animation, and 24 Gb of ram was not enough.

How do you plan to approach this?

plus crowd simulation is more than people wondering around, walking in none seance patterns, yes Autodesk is spending serious time developing this technology, (I am not fan of it because so far is not what they promise) but the behavior of the people make more sense than your previews, defining path, where the people should go or just hang out in one spot, is perfect because if you move the camera around you don't want to cross over any heads or make them walk in the wrong door.

So far anima is what we use here, and the one that works fine, it does have some issues, like stupid animations for character, waving hands and strange zombie position of some characters, but you can coordinate or choreograph in a very efficient way where they walk if they stop or not for a moment, if they stay seated, going up and down stairs and so.

Again Kudos on taking the initiative of try to develop something different, but particle system in Max I am not sure is s solid foundation for it, hopefully I am wrong

[Scott Schroeder]

Particle system is okay for general crowd wandering or milling about (or zombies chasing someone), but logical pathing is where it fails. You generally will get people who are blatantly walking towards a wall or another human, then change direction suddenly when they hit the blocking event. Most humans aren't that robotic, which is why crowd sims are extremely hard to develop correctly. You can't program free will, and if you do, please don't name it Skynet.

 

$500 bucks is a bit too pricey for what you are currently showing. Realistically, it's nothing groundbreaking and nothing that anyone who knows Pflow couldn't do in a weekend. I know you are still in the dev stage, but you have to put some wowzers in my trousers before I'd even consider paying your asking price.