Welcome to od|forum

Register now to gain access to all of our features. Once registered and logged in, you will be able to contribute to this site by submitting your own content or replying to existing content. You'll be able to customize your profile, receive reputation points as a reward for submitting content, while also communicating with other members via your own private inbox, plus much more! This message will be removed once you have signed in.

Jason

Houdini 9 Wish List

187 posts in this topic

Hey Guys,

Well we don't need perfectly simulated thought. The thing is the AI is useful for more than just Crowd Sims. Take the course in Siggraph 2004 on Bayesain Learning. They used the Bayesain learning system to make an effective automodeler for character modeling. Of course this could be applied to other modeling techinics. Also it would be used with Particles to kind of have directed effects in a totally different way. Or a realistic swarm of bees or birds. I know we have flocking which is great but still it would be interesting. It could be used in shading to drive shading networks and perhaps displacements. Of course it could be used for crowd sims. Maybe even for rigging. I don't know but the applications are certainly there. I am thinking simple functions or simple brians nothing complex enough to need a whole lot of computing.

Cheers,

Nate Nesler

Share this post


Link to post
Share on other sites
Hmm ... I just took two torus and ran it through the Cookie SOP. It seems to leave the original polygons to me.

23388[/snapback]

Would that perhaps be because a torus generates convex polys?

Try putting an edit sop on them and randomly pulling a few points around as you would if you were free-form modelling, then try it again. In very simple cases you can turn off pre-convex geometry and it still works, however in most practical situations doing that generates errors.

Also, after doing the cookie, with pre-convex off, if you then try and do a subdivide you get cracks in the geometry, even if you fuse first.

Share this post


Link to post
Share on other sites
HOLY geeked out madness Batman!

All we need now is for Stu to tell us how many frames the wipes are in the Star Wars movies. (again)

B)

-k

23391[/snapback]

You're dead to me.

Share this post


Link to post
Share on other sites
Also, after doing the cookie, with pre-convex off, if you then try and do a subdivide you get cracks in the geometry, even if you fuse first.

Did you try turning on Consolidate Edges? But yeah, it's kinda hard to do intersections with non-planar polygon's isn't it? :)

Share this post


Link to post
Share on other sites
They used the Bayesain learning system to make an effective automodeler for character modeling. 

I don't know much about this stuff but I'm sure you'd need a very very specific application to apply AI techniques to instead of providing a general solver, no? At the end of the day if you manage to come up with a specification for a general AI solver then you'd probably have to spend a ton more time trying to build a careful custom network around that general solver to make the solver effective, no?

How would you see this happen? Can you give us all a clear idea of the inputs and outputs of such nodes? Try and come up with a convincing pitch for this.

Personally I'd like to see some more "dumb-but-fundamental" tools - nothing so pie-in-the-sky as neural-nets. Things like pattern recognition, 2d tracking, 3d reconstruction, optical flow and more support for state machine tasks. I think with a few fundamental tools in COPs and POPs, we could possibly implement fully rendered (software or openGL) eyesight views from particles(agents) and, with some Computer Vision techniques, enable them to "see", then think, then act and importantly, remember - all according to some credo/profile/personality.

If all these tools are built with certain solutions in mind, I have no doubt we could bullt the most awesome crowd simulations in Houdini - especially with the help of OTLs.

I'm pretty sure there are some great ideas and application for AI in Houdini - so please share:)

Cheers,

Jason

Share this post


Link to post
Share on other sites
Did you try turning on Consolidate Edges? But yeah, it's kinda hard to do intersections with non-planar polygon's isn't it? :)

23406[/snapback]

Ah that's a new option to H8, just tried it in a simple test, it's good to have but didn't work in the case I tried. Good to see some improvements though.

Hard to do intersections with non-planar polys, yes, but no need to convex polys that aren't intersected, no?

Share this post


Link to post
Share on other sites
Hard to do intersections with non-planar polys, yes, but no need to convex polys that aren't intersected, no?

23432[/snapback]

Isn't that a little contradictory? You need to convex them to check for intersections in the first place.

Do you mean pre-convex everything (keeping a map of which polys were convexed), do the checks and then restore the original geometry then re-convex only the intersected ones?

Share this post


Link to post
Share on other sites
Ah that's a new option to H8, just tried it in a simple test, it's good to have but didn't work in the case I tried. Good to see some improvements though.

Hard to do intersections with non-planar polys, yes, but no need to convex polys that aren't intersected, no?

23432[/snapback]

Did you try also adjusting the tolerance?

As for non-planar poly's my guess is that you have to convex them all since it doesn't know which poly's need to be intersected ahead of time.

Share this post


Link to post
Share on other sites

Hey Jason,

I was looking at AI for more than just Crowds. In order to effective answer your questions though I think I would have to do more research in AI because I only know bits and pieces and not enough to really give a full length discussion on the topic. UGA has expressed some interest in having me work with their AI department in the programming side so who knows. Of course if I dive that deep in it then I will probably just get into the water myself so to speak. I know massive uses Neural Networks and such to build simple brains from nodes for characters to form what is called Artifical Life. Which is ummm very interesting. Artifical Life is different from Artifical Intellegence but both are formed from the basis of thought. So I was not thinking of trying to have a fully thinking cg character that we could form consertations with and one that had voice recognition etc. I don't really see that as all that useful to us in the 3D Film Arena it might have a stonger application for games. I was kind of looking at it from a stand point of Artifical Life, and to solve problems in 3D and to create more advanced control systems that are more directed. I think this is a topic I will come back to later when I have more knowledge on the subject matter.

Cheers,

Nate Nesler

Share this post


Link to post
Share on other sites
Isn't that a little contradictory? You need to convex them to check for intersections in the first place.

Do you mean pre-convex everything (keeping a map of which polys were convexed), do the checks and then restore the original geometry then re-convex only the intersected ones?

23440[/snapback]

Bingo!

Share this post


Link to post
Share on other sites

Like most people I'd like to see improvements to Mantra GI. It would be nice to make it a prman beater!

DOPs fluid and hair would be fantastic too!

BTW: Jason, you missed a bit! :D:blink:

Share this post


Link to post
Share on other sites
Like most people I'd like to see improvements to Mantra GI. It would be nice to make it a prman beater!

I consider the GI in Mantra a prman match right now. It's faster with a smaller memory footprint -and you can perform some clever point-cloud techniques sometimes and this brings it level with the fine irradiance cache control you have in prman, IMHO. I've done numerous speed/memory tests with complex models and Mantra wins out in speed and quality every time.

Do you have specific cases or scenes where you consider PrMan superior in this area? I'd love to know what you think is still outstanding wrt to GI in Mantra wrt Prman.

Share this post


Link to post
Share on other sites

Well, I was just visiting someone who uses PRman instead of Mantra mainly because Mantra lacks easy baking of each object or group's Occlusion. Apparently, PRman offers this. Also, again second hand info, the gather() statement in PRman is useful, which Mantra lacks.

This is all second hand info, I have no recent direct experience with PRman to verify or offer Mantra alternatives.

I was involved in writing a per-group baker when I was at R+H and it was not fast or fun :(

Cheers,

Peter B

I consider the GI in Mantra a prman match right now. It's faster with a smaller memory footprint -and you can perform some clever point-cloud techniques sometimes and this brings it level with the fine irradiance cache control you have in prman, IMHO.  I've done numerous speed/memory tests with complex models and Mantra wins out in speed and quality every time.

Do you have specific cases or scenes where you consider PrMan superior in this area? I'd love to know what you think is still outstanding wrt to GI in Mantra wrt Prman.

23482[/snapback]

Share this post


Link to post
Share on other sites
Well, I was just visiting someone who uses PRman instead of Mantra mainly because Mantra lacks easy baking of each object or group's Occlusion. Apparently, PRman offers this. Also, again second hand info, the gather() statement in PRman is useful, which Mantra lacks.

The gather construct is pretty useful thats for sure but not a big enough feature to deny the quality/speed/cost advantage of Mantra, IMO.

The baked occlusion in PrMan is very useful, I'll very much agree- I wish Mantra could insert values into custom caches to help with that type of thing too. However, practically speaking we've achieved really good results MUCH faster by performing a simple UVUnwrap and doing a shader bake (mantra -u) with Mantra. The renderman cache is huge and has artifacts that are hard to conquer. Also, working with unwrapped maps puts a lot of control in your hands because the data is in an image raster and not in a hard-to-manipulate data structure. You can perform any image-processing you like on the maps and they're small and efficent and easy for the renderer to filter. To only potentially awkward thing to manage is that, depending on how you've set up your scene, you might have to manage many of these maps.

Share this post


Link to post
Share on other sites

Hi guys,

This is my first post to the group. I'm very new to Houdini and infact animation domain itself. I'm just learning the ropes. I downloaded the eitire vislab(houdini video tutorials) site and am trying to learn it.

Animation is only my hobby. I mostly use lightwave and love/hate(sometimes) it. I've used Maya and XSI a bit but not much(max sucks in many ways than one). Coming to Houdini, it s one impressive peice of softwares. Firstly its completely non-destructive(or procedural) and the second is that its been around for more than 20 yrs now.

In some other thread in this forum, I saw a post of someone saying that "Houdini takes this, proceduralism aspect, too far.Like adding a Curve SOP to a set of points to create a curve.I'd rather draw a curve."

Being consistent with your philisophy is extremely important and thats exactly what houdini is doing.There are lots of things the other guys at Avid|Softimage, Alias, Discreet, maxon and Newtek can learn from SideFX(instead/friendly support) and Houdini.

Staying consistent with your thing is one aspect. But the fact that Houdini is not king itself is criminal SideFX part.Nothing against the sidefx guys.But why are they not building a wrapper UI through which can do the same like drawing a curve, i.e. a wrapper UI which will help you build the network in the background while you build stuff interactively. I'm sure this can be done on top Houdini's power house engine.Why can't they do this? Freedom and Choice is both a power and responsibility. Hopudini provides both. But not make it a bit more task-friendly instead of project/design-friendly alone. The approach Houdini takes is admitedly the best there is but why not refine it a bit more. IMHO SideFX has to catch up with times.

One more aspect is that, if you guys love/dote Houdini so much, then why can't you write a decent review about it asa a new version comes out, especially the veterans who have migrated from other packages like 3dsmax,maya,sofftimage|xsi,softimage|3d or anything else out there? Most of the reviews out there are written by half-baked or biased and written by people who are either very new to houdini or biased(total/partial) towards other packages or against houdini. If Peter Bowmar and Will Cunningham can write a book, why not a review? Look at the way Ed Harris covers most aspects of XSI asa a new version comes out.

I have a request for you guys, especially the Houdini experts who have migrated from other packages like XSI or maya or lightwave.Please do write a review. If SideFX pays you or you do this for the Houdini community or for yourslef, the end result is that Houdini user base may grow in number because of what you've done. You don't have to necessarily compare houdini with other packages but since you are familiar with other packages, you might be able to put in the best of both worlds.

Pleaseeeeeeeeeeeeeeeeeeee............................

Maybe make it intuitive? Primary and secondary and really, the whole Twist SOP, are just the most unintuitive things ever. Please see Lightwave's modeller for intuitive modelling deformations like this. Extremely interactive and intuitive.

Cheers,

Peter B

23013[/snapback]

Share this post


Link to post
Share on other sites

Here's my wish list for h9:

Mantra: (haven't used intensly for some time, but I do believe these are still outstanding)

- gather() construct a la prman. With the ability to pass parameters to shaders before they are evaluated. this would make writing shaders x100 easier!

- ability to write out point cloud data/occulsion/irradiance caches on a per object level. this is also of vital importance. Similar to prman's:

Attribute "irradiance" "filemode" "w" "handle" "$MAPS/occulsion_cache/$OS.icf"

ie, this can be specified either in the shader OR in the rib stream. Very very useful for complex scenes where you want to reuse 90% of the static points and only recalculate the objects that are moving.

- is anti-aliasing finally "fixed"?

- speed up occulsion/irradiance. Does mantra have importance sampling yet?

Assets:

- rewrite how vop definitions are stored. Right now, there is a severe architectural problem with DA's in general. There is currently two ways of creating them. The VOP way, and the rest of the system. The VOP method is correct, embed the definition within the asset itself. The other way, a seperate definition in the VOP context is an unnecessary extra step. This means that it's hard to make a self contained asset with it's own sub definitions - as having several of them also means you have several definitions of it's sub assets. A real mess! This is my #1 Houdini 9 wish.

- asset manager. need a rpm style versioning with auto dependencies. I'm not talking about a full CVS style versioning which I think is just not needed. Just a more granular asset A version 1.3 needs assets B(v.2) C(v2.3) & D(v3.4) in order for it to work. This is not a biggie to write - just some low-level management stuff. DA's are a great start - a proper asset mangement system would add x100 value!

Cops:

- make it work

- 2d tracking (cmon sesi, this has been outstanding for >5 years...) just buy it in!

Regards,

Paul

Share this post


Link to post
Share on other sites
i.e. a wrapper UI which will help you build the network in the background while you build stuff interactively. I'm sure this can be done on top Houdini's power house engine.Why can't they do this?

:huh: i guess i don't understand.. can you not just close your network pane and work directly through the view port like any other packages? Can you not just select "Curve" from the menu or toolbar in your viewer of choice and just start plopping down, removing, and shifting points? How is this different from package X? Granted, most people work through the network just as myself, but that doesn't mean you have to.

Share this post


Link to post
Share on other sites
This is my first post to the group. I'm very new to Houdini and infact animation domain itself. I'm just learning the ropes. I downloaded the eitire vislab(houdini video tutorials) site and am trying to learn it.

Welcome to Houdini! :) The old vislab tutorials are good but they're more task oriented. You probably want to check out the videos on Houdini over at 3dbuzz.com as well if you're just starting.

Share this post


Link to post
Share on other sites

Can we ask for removal of things to be in the poll?

I think it's about time custom panels were banished to the far corners of the earth and any and all saveable functionality rolled into HDA's.

M

Share this post


Link to post
Share on other sites

Speed,

Until I built something serious I thought It's fast in most cases but it's not.

Especially in deformations on complex > 300k geo (not being a default sphere :) ) and standard operations like selection, switch display mode (repeating after Jason) so on.

Lattice, edit sop, delete....all works painfully slow.

This is a very important thing since the secnes are huge nowadays and some other software can handle them easily.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now