hi folks,

as promised I’m going to spoil the fun by throwing in some facts, figures and observations from Cloth Lab (a special branch from Muppets Labs Inc).

At first I loaded just Vicky 4.2 to test her leg, from the issue on body parts intersecting each other, and on poses being feasable in real life. My finding is that the knee can bend up to 105* in a very natural way, while up to 120* the upper and lower leg start squeezing in the knee-pit but not having an unnatural result. Above 120* (the limit is 145*) the intersection starts limiting proper results, while my own legs start feeling rather uncomfortable with bends from 120 to 135. I can’t make more myself. So, I would like to consider every bend below 105 as relaxed, up to 120 as tight and over that as extreme and avoidable. The 145* limit is good for Android Andy, but that’s it.

Then I loaded JeanZ from 3DWizard, used it as conforming, and made the same bends. Again, up to 105* offered no problems but close to 120 the pose started to be unnatural as the thickness of the cloth made the self-intersection start earlier. As could be expected, the conforming clothes hardly folded and wrinkled which produced a very unnatural result, like constraining dynamic pants from hip to below the knee. It’s like a panty-hose.

Then I turned the conforming JeanZ into dynamic, using the build in Wardrobe Wizard conversion script, choosing for high resolution and keeping all morphs. This pumped to 5Mb obj up to a 60Mb pp2 file. The first trial run made all the bottons and nails pop off, so I made a soft-decoration group for them (constraining them to the cloth) and constrained a strip from the belt-area to Vicky’s hips, to prevent the JeanZ from falling off. Setting the drape to 5 frames, everything else was kept to default. One leg moved in a 30 frame animation from straight to a 90* bend upper leg and a 120* bend knee.

While frames took between 5 and 10 mins each to simulate and the 4-core CPU fluctuated between 25 and 75% usage on Poser alone, the pants started folding in the crotch area in a very natural way, and the knee area started folding later and stayed very natural up to 105*. Up to 120* the sharp folding as raised by Reinhold Pupu (dolherin) originally became visible. Is was clear that the cloth was pulled into the kneepit and started self-intersecting for a bit.

Then I redid the scene completely, and started to convert the original JeanZ to a lowres clothified prop without any morphs, using the same WW script and the usual cloth room steps. The 5Mb obj became a 12Mb pp2. Same setting for a soft-deco group and constraints, default cloth parameters again, frames again took 5 to 10 min each and the CPU ran 25-75% and even more on my 4-core machine, which means it has 1 to 4 threads active simultaneously. I’m on Poser 8, Vista-32 by the way. See my post on multi-threading above.

Like in hires, crotch and knee folding started right away and looked pretty natural up to 80* (frame 20/30). Beyond that, the pants became self-intersecting and not being able to make the fold, causing the issue that gave rise to this thread. From frame 24 (96* bend) results got ugly.

So the issue is a combination of self-intersection of the pants leg (not Vickys leg) due to the pose and a lowres pants mesh. A hires mesh suffers from self-intersection too, but at a rather larger angle (higher frame nr), and therefore can handle more extreme poses.

Lowres calculating in about similar times as hires did not supprise me that much. In the first place, the pants suffered from a “naughty vertex” which was off all the time and might have caused the long times in the first place. Without it, times might even have been 30 sec for lowres and 60 for hires, which is a difference, but I can’t tell now. Second, cloth simulation is a collision detection between two meshes, and reducing pants only but keeping Vicky intact might not have made a huge effect.

At last, I ran a simulation with all cloth parameters at 1/10th of their default values. This results in jeans made from a mixture of latex, silk and lace, and technically gave all vertices more freedom to move and settle. The result was a 5 times longer calculation time (2000 sec/frame instead of 400) and pants being torn apart in small areas or showing other artifacts not seen before. Not very recommended. Cloth parameters just should get as close to reality as possible, apparently.

Back to the original issue. Reinhold used the fold and artifact issue for raising the question on future development of the dynamic cloth features. No-one knows, it’s all a matter of time and money, or: user-base, user-request and competition. The latter has become more severe, from Daz Studio (half the price), Carrara (double the price) and Blender (free) all ramping up functionality fast. On the other end, most studio tools (Max, Maya, SoftImage, Cinema4D, ...) dropped their low-price varieties.

There is a thread in the forum on user requests, like “what about version 9” or so. It’s all about user interface and speed, library handling and bugs / stability, lighting and mesh decimation / subdivision. I just waded through all pages of that thread but I hardly ran into requests on dynamic hair, cloth or alike. Are we on a lonely island on Poser planet? On the other hand, improving dynamic cloth is not on top of my priority list either.

This does not imply that the dynamics cannot be improved upon. Although the current simulation works fine in non-extreme conditions, the paper referred to by ice-boy proves that revised algorithms can be utilized for handling crushed cloth in cramped knee- and armpits, and shirts being rubbed. This still leaves the question whether such a thing should get priority in the further development of a $200 program that – to my knowledge – has never been used for any 10+ minutes movie, music-clip, price-winning short, artistic DVD production or TV commercial. IMHO, it should not. It is on my wish list, but not on top. This implies that for stills, I’ve got to fix things in post. For 3D-cartoony animations I’ll have to find tweaks, and for photoreal shorts I’ve got to avoid extremes in front of the camera. Fine to me.

That’s all, for now.
Happy Posing.

hi all,

while my promised tests are still running (current settings require over 30 mins a frame now), I'm going to spoil the multi threading debate in this dedicated post. Sorry.

Multi threading is a limited venue. It might speed up a process in a lineair way, doubling the threads cuts the time in half, but at the cost of quadrupling overhead due to interaction between threads or thread results. so there always is some upper limit where overhead grows faster than speed increase.

In the final rendering process, this limit is quite high, because calculating the image pixel color from polys, materials and lights does not effect those polys etc in return. It's a one way non-looping calculation. This is why GPU processing using over 1000 CUDA processors in your nVidia card still can be a gain (not for Poser yet).

In cloth simulation however, moving a vertex does effect the calculation of other vertices thanks to stretching and sheering, and this looping calculation limits the multi-threading severely at the algorithmic level. Currently 2 to 4 simultaneous threads might be the limit. A complete new algorithm (be my guest) might bring us 8, but thats really it.

And while running all my tests, Taskmanager tells me that Poser is using 25-75% of my 4-core machine. That is: 1 to 3 threads in parallel on cloth simulation, part of the time. So up till now my conclusion is that the Poser cloth sim process (just P8 in Vista-32) is as multi-threaded as can be, for a $200 program.

Of course I understand that people like to have the functionality of a $20.000 pack, and that other people like to pay only $2 for it. I'm happy with what I've got now.

And indeed, for calculating a simulation a very fast 2-core CPU is better than a slow 6-pack. For rendering it's the reverse. Life is one big trade-off.

More results to follow soon, be prepared...

hi Ice boy,

there is. Normally used for a dynamic jacket over a dynamic shirt.

Just put on the two pieces of clothes, or the shirt first. Then create simulation 1, and have the first one (the shirt) collide to the character (say Vicky). Run the sim. Then create simulation 2, and have the second one (the jacket) collide to the first (shirt) instead of the character. Run that sim too.

Then render the animation, containing both sims.

But I don't agree you cannot calculate on advanced clothing. It's not that difficult to have a linnen T-shirt with leather bands stitched onto it, and so on. You can assign linnen and leather different material properties, like stretch resistance, density, air-damping and you name it. Same for double stiched pockets and the like. With just an advanced definition of the cloth, one sim does it all.

For example, I'm just testing right now on the issues above, using Vicky and JeanZ, all morphs included (say 60Mb). the JeanZ contain double stitched pockets front and back, belt loops, a leather belt, iron buttons (coming off right now, oops) and whatever. Due to some poke throughs, simulation requires 5 to 10 mins a frame in basic setup. Will be improved upon.

I guess studios use low-res versions first in order to get the underlying movements right, the poses, the flows, the muscle bulging. Then they shift to hi-res. Just assuming, that is.

regards

hi bevans84,

see where you're coming from. Point is that constraining pants from hip to below the knee doesn't leave much dynamic, and as you said yourself, doesn't give much wrinkles and folds either. Therefore, it doesn't look like jeans pants to me, while constraining the belt should be enough.

I'm afraid this doesn't help Reinhold, as "take a helicopter" is not the answer one is looking for on the question "faster path by car". It's faster indeed, true, but the car got lost in the process.

Your notes are well appreciated though. Let's go for the folds. This is an interesting thread, isn't it.

I agree, and there are more obstacles to take. Like the fact that serious scenes in Poser and mainly Vue blow up to say 4Gb vertex/poly info before rendering kicks off. I shouldn't know how to put this into GPU's but exporting to COLADA sure is not the way.

But in the meantime I do hope that the technology will evolve, and the CUDA GPU's on the nVidia cards will get utilized by those render-intensive 3D engines.

In the meantime, Octane does have some advantages for realtime high quality rendering in various situations, like automotive commercials, architectural flyby's and hi-end gaming. And it does surpass OpenGL when 3d painting / texturing and 3D sculpting using a tablet (ZBrush style).

This makes it a good proof-of-concept. I like the concept, and enjoy the proof.

to bagginsbill and englishbob,

you know, I'm just not sure. Because when upper and lower leg intersect at the kneebend, there won't be any room left for the cloth inbetween, and the simulation would make the cloth flow around it like in real life, unless the pants are so tight that a clash cannot be prevented. Relaxing the leg such that the intersection goes away would not make the pants cloth flow correctly. This is why I wonder whether any intersection can be an underlying cause.

So I'm curious to your considerations, why you think it's so relevant.

And I haven't seen the model without the pants (yet, perhaps Reinhold can show us), so I can't tell. I guess I'll have to fire up my Cloth Testing Lab :-) for this weekend, then, to find out.

To be continued...

Hi Reinhold,

I’ll try to help you a step further, as this is an interesting and far from trivial situation. This might turn this thread into a mini tutorial, but who cares. The main questions to solve are:

• is the situation feasible in reality
• is 3D able to follow reality close enough
• if not, what else?

Reality check

This slender person without heavy leg muscles might be able to bend the legs up to 120*, but be aware that lower and upper leg skin will start touching behind the knee from 90* on. Bending the leg with elongate the path over the knee, hip to toe, with about a hand length (say 15-20 cm), and will shorten the path at the back, from buttocks to heel, with about the same amount. As a result, the pants will raise at the ankles, and an awful lot of cloth has to be wrinkled away behind the knees. Looking at my own pants, four serious folds seems to be the minimum to get the job done, and folding starts halfway my upper leg already.

If the jeans are tight or thick the pants will have problems making all the folds and wrinkles, and will actually hamper the move and pose. The cloth will get crushed behind the knee, and things become uncomfortable in due time. 

When looking at your image, I only see one sharp fold behind the knee, and that’s it. It seems like high fold- and stretch resistance to me, thick lumberjack jeans or even leather like. The pants seem tight around the upper leg too, so the pants get sucked into the tight area behind the knee instead of being able to avoid it.

So my first tests would be to widen the hose a bit, relax the cloth parameters and actually animate the bending of the leg while sitting, to enable the simulation to get the cloth and folds in place. Just a few frames will do, and you might increase the number of steps per frame to 5 or 10 (instead of the default 2). This will increase simulation time a bit, of course.

3D ability check

As pointed out in posts above, Poser does support hard bodies only, and everything soft has to be emulated using morphs and other distortions. Cloth simulation is not different, it’s just a way to derive those distortions but the cloth remains a hard body too. Objects passing through each other can be prevented by switching on collision detection, but that’s about it.

First, cloth simulation offers a basic collision detection plus three extra features. Use them all, they only give real longer calculation times in case they matter. Which is in cramped conditions like this, and when making wild moves or extreme poses while the cloth has a relative low poly density. The simulation offset give the cloth its thickness, that is: when set to say 1 mm it will remain this distance between cloth and skin and between cloth and itself. Anything within this distance will be considered a collision.

In your case, a decent flow of cloth behind the knee requires 3 mm: 1 for the upper leg, 1 for the lower leg and 1 for the cloth bending and folding onto itself. The issue in your case is: when the pants are so tight that the cloth is sucked into the tight area between upper and lower leg, this room may not be available. So reducing the simulation offset might help in cramped situations.

Second, is the cloth 3D mesh able to make the folds and wrinkles? If you are relatively low in polys in that area, two neighboring ones are folded about on top of each other like making a sharp fold in a sheet of paper. And since smoothing of a mesh stops just before 90* bends, those folds will come out quite sharp. The solution, as pointed out in posts above, is to crank up the poly density at least a tenfold (order of magnitude). Just doubling won’t do I guess.

What else?

Poser was designed for artists to get the main things right, and work from that to a sculpt, a real (analogue) painting or a digital illustration. Most people work like that, and make the final image in post improving on everything a tool like Poser can’t offer. Wrinkles, lighting details, color dynamics, camera features like depth of field, and so on. 

So, when the objects and tools fall short (and they will, even in version 900), nothing stops you from painting the folds by hand. Or to trade a few hundred bucks Poser for a few thousands bucks feature film tool like Maya or so. Well, your wallet may.

Hope this helps.

Ronald

hi all,

the tutorial is in the Renderosity Tutorials section too, now. Actually it's in some Poser section thereof, but who cares.

have fun, and render.

update: the tutorial is in the Renderosity Tutorial section too, now.

Comments anyone?

In my opinion, the Poser dynamic clothing is very very good, and I do know my way around animation, cloth and fluid dynamics and the like. The dynamics follow physics laws quite well.

A few remarks though.

1. the simulation settings offer some checkmarks to elaborate on the default vertex-to-poly collission detection. The definition of the dynamics also offer options to stick parts of dresses etc to the body. One does need and understand those things, especially for handling extreme cases like tight or thick clothing, or wild moves in animation. I did a thread on this some time ago.

2. the cloth might need some parameter adjustments, to define its structure, internal frictions and resistances and response to forces. The simulations assume a clean atmosphere of medium density, medium humidity and so on, wind forces can be added and perform according to physics laws. Gravity is according to average Earth surface conditions. Other conditions, like rain, water absorbtion, underwater cloth movements, exceeding the sound barrier and dancing on the moon or Jupiter are not supported, but can be emulated by changing or animating the cloth defining parameters.

3. figures movements should stick within limits or even more restricted to avoid unnatural positioning of limbs and body elements. Age, condition, obeseness and clothing itself ! are known limiting factors. One cannot make sharp leg bends or perform Swans Lake in a tight hose of thick leather. You also cannot break world records at 100 mtr sprint wearing a royal robe or evening gown. This means that when simulations go wild even after setting the right parameters, you're usually trying to make a move or take a pose which is highly questionable in real life.

In my personal opinion, all these are valuable assets, and I am quite happy with a feature which prevents me from treating carpet for lace or so, and questions my animations given clothing. But dynamics in all its facets it quite a complex area, so I promiss to make a thorough advanced level tutorial on this (before the end of this year). So, when all it set properly and simulation still cannot follow your poses, you might resonsider the poses instead of the dynamics. Again, in my opinion.

My only annoyance is that the gravity in Cloth dynamics differs from the gravity Python script somewhere in the menu. So when you have a ball and drop it by the gravity script, and you have a very dense piece of cloth with ZERO air resistance, and drop it by cloth simulation from the same position, they are not falling equally fast as they should do. One of the algoritms is correct for 30 fps but well off for 25, the other takes a middle position and is slightly off for both. As a result, dropping a table with tablecloth result in physically incorrect results.

I already filed this in the thread on Poser 9 fetures.

I'm open to all questions and suggestions, to deal with in the tutorial.

Regards.

I like Octane (and the Cubix system expander, see www.cubixgpu.com), but with all those external renderers: you have to redo camera, lighting and (advanced) materials. What's your experience on this?

okay,

in another thread I found a link into Smith Micro, on solving Poser weird behavior. One never knows.

hi all,

in case of memory issues on 32-bit Windows, I did an extensive tutorial on it recently. It's in the Tutorials section. 

Why reducing priority (I use Taskmanager for it, great for long renders) or separating processing should help, I don't know but it doesn't hurt either. 

My Poser (8.3 in Vista 32) has the same symptoms, and crashes when switching to another Room. Switching from cloth room to material room is a favorite crasher, switching from cloth room to hair room and then to material room avoids crashes, and so on. Has nothing to do with memory or CPU, has nothing to do with scene complexity, getting the latest video-driver sometimes helps for a while.

Poser is just not 100% stable, so indeed, save frequently and before rendering or switching rooms in any case. Things will improve, given time. And get worse, in between. Software like Poser is more complex than the Space Shuttle, you know. Windows is more complex than Houston Space Control, and Microsoft has more programmers and performs more testing than NASA. So just let's enjoy the miracle of Poser ad OS's working together.

Have fun.

hi Andrew,

glad you fixed it. And that you found Sound in Control Panel, as I tried to point you at in the first place. Sorry for putting you on the wrong foot by translating my Dutch OS verbs incorrectly, it's Control Panel indeed (and not Configuration Screen). Thanks Sams3d for the screenshots and details.

Ronald

Sorry to hear, my condolances.

For some it might help to turn mood and emotions into something creative. You'll find your way.

Ronald