Animating Sub-d Character Super Slow

[goldrake]

i just finished modeling and rigging my first sub-d human character. now i'm starting to animate, and the lag is unbearable! i'm on a quad core workstation with 4GB of RAM running maya 8.5, so i thought it would be significantly more responsive.

i've tried setting the sub-d display smoothness value to zero, and that helps a tiny bit, but basically it's still impossible to preview even the simplest movement without a playblast. is this normal??

i have nothing visible in the scene other than my character, which is not that detailed. i'm not sure how to tell how detailed a sub-d model is (is there a way to count faces?), but my original poly model (i converted the poly to sub-d) was about 1800 faces. the character is untextured and unlit.

i tried animating the ball joint... just a simple toe wiggle... and i was getting 1.4 fps!! this can't be right, can it??

any help would be appreciated. thank you.

[joojaa]

did you bind the 0 level or all levels?

[theelephantfactory]

i've been in situations like this, and although it might not be ideal, i've gotten around it by making a low res version of the character out of primitives and just parenting the primitives to the proper joints. of course this won't work for facial animation, but for the body, it works fine and should be nice and fast.

[goldrake]

thanks for the tip, elephant. i guess that will be my last resort.

joojaa, i'm not sure which level i bound... how can i tell? my display smoothness was set to 3 when i selected the geometry and bound it... does that answer your question? how do i bind to level 0?

[goldrake]

oh wait, i think i understand what you mean. i bound level 0. i am pretty certain because only my coarsest vertices have weights assigned to them.

[goldrake]

ok... did some research and came up with a few solutions.

1) BIND SUBD POLY CAGE
Joojaa, this is actually something you suggested in THIS EARLIER POST. Basically, I bound the subd poly cage to the skeleton, gave the cage a shader so i could see it, and hid my subd model by unchecking subd surfaces from my Show menu. With this solution, I was able to get 7fps, which is a lot better than 1.4fps, but still too slow in my opinion.

2) CREATE POLY WRAP DEFORMER
Joojaa, you were discouraging this approach in the post referenced above, but I was nonetheless able to get 22-24fps with this solution. Perhaps it's less elegant come render time, but for timeline playback, it seems like the way to go, and as long as the renders come out fine, I don't mind. But please let me know if you disagree, as this is all new to me, so I don't really know what's best. In any case, I went back to the poly version of my model (the last version before converting to subds, but i guess i could have also created a poly from the subd version), lined it up with my subd model, bound the poly model to the skeleton, and then used to poly model as a wrap deformer for the subd model. The subd model has to be hidden obviously in order to get real time playback. Just turn it back on (and the wrap deformer poly off) for rendering.

Curious to hear about other solutions now... or just people's thoughts on the above.

[goldrake]

i guess, if possible, i would like to bind the subd poly cage to the skeleton as opposed to using a poly wrap deformer... just because it strikes me as the better approach. but i just can't seem to get the same performance at the timeline level.

as a test, i tried selecting the subd poly cage, duplicating it so it was a freestanding polygonal object with no history or anything, and binding that to the skeleton. so this is just a 3700 face polygon bound to a skeleton... but still the performance was 7fps. weird.

i compared this result with my original polygon model (before conversion to subd). that original poly model has 3600 faces (i stated 1800 above, but that was wrong), so only 100 fewer faces. and yet i get 22-24fps! why would that be? is the duplicated poly cage not a real polygon?

(is anyone listening or am i just having a conversation with myself? )

[joojaa]

QUOTE(Goldrake @ 11/03/07, 03:28 AM) [snapback]275853[/snapback]
2) CREATE POLY WRAP DEFORMER
Joojaa, you were discouraging this approach in the post referenced above, but I was nonetheless able to get 22-24fps with this solution. Perhaps it's less elegant come render time, but for timeline playback, it seems like the way to go, and as

Yes but you can get 48-60 fps withouy the wrap.

Its a stupid indirection imagine:

a+b+c-c=a+b

So what your doing is c-c here. You keep telling the computer to do thsi over and over and over again 24 times a second now since c and -c are atleast as heavy operations as a+b your actually spending 1/2 your time doing that part. Secondly you can get into trouble with this. Altough the wrap was redesigned to account for this (i think the idea was they gave up on the notion their users are smart, in essence raising hands and saying fine have it your way) sit still not the tool for it.

Remember that the polycage is the same as your 0 level points, so why would you spatially search for a corresponding point and then move thet accordingly because you already have that map. Incidenttaly going into polygon mode makes a inetersting a-2a+b+2a roundup. But thst just because you painted the unevenness on allready (and yes if you had yet to make the un evveness pait youd end up with a a+b kind of situation which obviously is the fastest)

However the 0 level points bound should be as fast as the wrap method (but you loose the ability to use the polycage as proxy). Its just teht people sometimes build things upside down and they end up building history on history, wich is why its important to make the uneveness modeling at right point of time.

Offcourse theres also a very big probability your mesh is not really subd optimal.

PS: remember that your killer computer is NOT fast enough for most serious jobs. So don't waste its resources. The problem here is that you can get it to work no trouble there fine. But consider when you really end up problems even with this method your even more far gone into this and i bet your ass off your be on REAL deadline by then and its quite considerable feat to unlearn all you learnt allready. This si allmost impossible under time presure.

[goldrake]

Hi Joojaa,

thanks for your reply. i understand why theoretically the wrap deformer is inefficient, and like I said before, I really want to do this the "right" (your) way, but I can't seem to get fast results except with the wrap deformer, and I think that may be because I don't entirely understand what I'm supposed to be doing.

QUOTE
you can get 48-60 fps withouy the wrap.

could you kindly explain step-by-step how to achieve these speeds? i modeled in polys, converted to subds, now what do i do? do i bind the subd poly cage? do i bind level 0 of the subd? i tried both of these methods and both were slow.

also, there were a few things you wrote that i didn't understand...

QUOTE
Remember that the polycage is the same as your 0 level points, so why would you spatially search for a corresponding point and then move thet accordingly because you already have that map.

are you saying binding level 0 is the same as binding the subd poly cage? are you saying i shouldn't bind the subd poly cage?

QUOTE
Incidenttaly going into polygon mode makes a inetersting a-2a+b+2a roundup. But thst just because you painted the unevenness on allready (and yes if you had yet to make the un evveness pait youd end up with a a+b kind of situation which obviously is the fastest)

what do you mean by "painting the unevenness"? are you taking about the fact that after i converted to subds, i made the model asymmetrical?

QUOTE
Its just teht people sometimes build things upside down and they end up building history on history, wich is why its important to make the uneveness modeling at right point of time.

What would be the right point in time in my case? Did I do it wrong?

QUOTE
Offcourse theres also a very big probability your mesh is not really subd optimal.

Could you explain a bit what you mean by that?

Sorry for all the questions, it's just that you're the only person giving me feedback on this, and I don't entirely understand everything you wrote. I hope you have time to reply. Thank you.

[goldrake]

sorry, one last question... when i convert my poly model to subds, am i supposed to click on the check box "Proxy object" in the conversion options? because i didn't do that initially, so i went back and tried it again both ways, checked and unchecked.

with proxy checked, i was getting 70FPS! without it checked, i was getting 22FPS (22FPS is obviously a lot faster than 7FPS i was getting when i bound the poly cage earlier, but i'm assuming that's because i haven't creased edges and done additional subd modeling). in both cases, i was binding the subd poly cage.

if checking the proxy checkbox is the solution, that sucks because i did a lot of additional modeling in subd mode...

[goldrake]

ok, i seem to have found a workaround of sorts...

EXPERIMENT 1: BIND NON-PROXY SUBD LEVEL 0
1. take original poly model
2. convert to subd (proxy option UNCHECKED)
3. bind zero level subd
RESULT: 2 FPS

EXPERIMENT 2: BIND NON-PROXY SUBD POLY CAGE
1. take original poly model
2. convert to subd (proxy option UNCHECKED)
3. bind subd poly cage
4. uncheck subd surfaces from Show menu
RESULT: 22 FPS

EXPERIMENT 3: BIND PROXY SUBD POLY CAGE
1. take original poly model
2. convert to subd (proxy option CHECKED)
3. bind subd poly cage
4. uncheck subd surfaces from Show menu
RESULT: 70 FPS

in view of the results above, i converted my original polygon model twice -- once with proxy off (non-proxy subd) and once with proxy on (proxy subd) -- and opened up the hypergraph to see what it was about selecting proxy that made all the difference. that's when i noticed that the non-proxy model had 12 subdBlindData nodes connected to the non-proxy subd poly cage.

just for the hell of it, i tried deleting the blindData nodes... they were the only visible difference in the hypergraph, so i figured i'd give it a try so long as i was experimenting. i then bound the non-proxy subd poly cage to the skeleton, and... amazingly enough, 70 FPS!!

now, i'm not entirely sure that this is a legitimate solution. i tried researching blindData, and this is what i found:

QUOTE
Blind data is information stored with polygons which is not used by Maya in any way, but is useful to the platform to which you export the polygons, such as a an interactive games console.

For example, when you use Maya to create content for interactive game levels you can use blind data to specify which faces of the level are "solid" or "permeable" to the character, or which faces on a polygon mesh are lava and hurt the character, and so on.

this character is being used for a film, not an interactive game, so i'm guessing i can go ahead and delete the blindData safely??

the only alternative i can think of is to start over from my original poly model, and convert to subds with proxy checked... but frankly, that's not a viable option at this stage, as i made a ton of changes to the model in subd mode and it would really, really suck to have to redo all of that.

anyway, let me know what you think when you have a chance. thank you again.

[goldrake]

wish i had seen this earlier... DOH!

SUBD SLOWNESS

anyway, nice to know i'm on the right track.

[joojaa]

The blind data nodes in this case contain the hierarchical edits you've made (they dont mean anything for your poly but do for subd), ie the unevenness you've painted. However reintegrating this data is a bit slow since what actually heppens is yoru doing the same thing 2 times.

Now the 0 level bind should also be 70 fps provided thet you delete the extra edit levels, you can then reintroduce them later. Remeber speed i directly propotional to the amountg of points you have in use. And that is NOT the amount of polys in your base cage but the amount of points COMBINED between all your levels! which is obviously quite a lot more, in your case 12 levels making your point cout maximum a ataggering 1800^4-7 wich oviously is a LOT more then your bargain for)

However theres a trick you could use. You dont actually need the hierarchical edits you've made untill you render right? Well then just connect the bastards to a custom slot so you can find them later and unconnect them form your mesh input. You can then reconnect them before you go back to render phase prefeerably with prerender script and take the 22 fps impact at render time only. (you can also disconnect them partially if you need to se some of the uneveness)

PS: a lot of problems arise because people dont watch their hyper graph. Wich is maya rigging tool NUMERO uno.

[goldrake]

Crap... I totally thought I had found the solution, but you are completely right: delete the blind data, and I lose the tweaks I made after converting to subds. I didn't see it initially because I didn't close and reopen the file. I guess the effects of the deleted blind data don't show up until you relaunch the file.

Anyway, I am now trying to figure out the workaround you suggested, but I had a few questions.

QUOTE
connect the bastards to a custom slot so you can find them later and unconnect them form your mesh input.

What do you mean by "custom slot"? I tried a null node, but I wasn't sure what to connect the blind data to. Which attribute? Right now, each blind data node is connected as follows: blindData[X].message --> bodyShapeHistPoly.blindDataNode[X-1].

QUOTE
You can then reconnect them before you go back to render phase prefeerably with prerender script and take the 22 fps impact at render time only.

my only concern here is that the blind data disappears when I bind the subd poly cage... in other words, it's only visible in the hypergraph BEFORE i bind. so if i select it, move it to a custom slot, disconnect it from the cage, and then bind the cage... will i then be able to reconnect it just before render?

[goldrake]

Here's what I did...

EXPERIMENT 1: BEFORE BINDING TO SKELETON
1. break blindData connection to UNBOUND subd poly cage.
2. subd reverts to original state, prior to creasing and tweaking.
3. reconnect blindData to UNBOUND subd poly cage.
4. creasing and tweaking is restored... so far so good!

HOWEVER...

EXPERIMENT 2: BOUND TO SKELETON
1. break blindData connection to UNBOUND subd poly cage.
2. subd reverts to original state, prior to creasing and tweaking.
3. bind subd poly cage to skeleton.
4. reconnect blindData to BOUND subd poly cage.
4. creasing and tweaking is NOT restored! blindData seems to have no impact on model.

it seems that once you bind, you can no longer modify the geometry... which is a problem, because i would obviously need to bind the geometry in order to animate it.

is there another solution perhaps that i didn't consider, joojaa?

also, what is it about my workflow (model in polys, convert to subds, model some more in subds, rig) that brought about this problem. i have never worked with subds before, but it seems to me that my workflow is pretty standard. i guess if i converted to subds without any additional modeling, this could have been avoided, but what's wrong with continuing to model at the subd level?

[goldrake]

joojaa... you are a genius! i figured out how to reconnect the blindData nodes. they don't actually disappear when i bind the poly subd cage. they are just transferred to a different node. prior to binding they are connected to "subdBodyShapeHistPoly". after binding, they move further upstream to "subdBodyShapeHistPolyOrig".

so basically, just to recap:

1. i select the subd poly cage (without deleting the blindData)
2. i bind it to the skeleton
3. i then locate subdBodyShapeHistPolyOrig
4. i disconnect the blindData
5. i save and reopen the file
6. once the file is reopened, my subd shape node has lost all creases (but that's ok, because i can get them back later)
7. i hide my subd
8. i animate my skinned subd poly cage (with no blind data, i get 80fps playback)
9. once i am happy with my animation, i reconnect the blind data
10. i save my file and reopen it
11. once the file is reopened, i turn my subds back to visible... i should now have all my creases back
12. i hide my subd poly cage, and render!

i'm going to run some more tests to make sure this works well. i guess it's a bit awkward as a workflow, but if i write a few scripts, it won't be too bad.

are you sure there are no negative consequences to disconnecting and reconnecting the blind data? seems to to work, but i just wanted to make sure...

[joojaa]

Adda multi attribute and connect the blind datas there. then just make a script that cheks every thatnamed multi in a poly shape and swaps those connectsions to inputs. and one that does the same in revrese.

Done, but i was thinking maybe its enough just to set the nodes states to blocking. then you wouldnt need to reconnect them

[goldrake]

i'll try that and report back. thanks again.

in the meantime...
QUOTE
are you sure there are no negative consequences to disconnecting and reconnecting the blind data? seems to to work, but i just wanted to make sure...

QUOTE
also, what is it about my workflow (model in polys, convert to subds, model some more in subds, rig) that brought about this problem. i have never worked with subds before, but it seems to me that my workflow is pretty standard. i guess if i converted to subds without any additional modeling, this could have been avoided, but what's wrong with continuing to model at the subd level?

what are your thoughts on these two questions?

[joojaa]

QUOTE(Goldrake @ 11/07/07, 07:15 PM) [snapback]276151[/snapback]
are you sure there are no negative consequences to disconnecting and reconnecting the blind data? seems to to work, but i just wanted to make sure...

Well in the past maya could under some circumstances not resolve this, but as long as its connected in the last node its there. However if you leave the nodes freestanding in maya scene without any connections they might disapear.

QUOTE(Goldrake @ 11/07/07, 07:15 PM) [snapback]276151[/snapback]
also, what is it about my workflow (model in polys, convert to subds, model some more in subds, rig) that brought about this problem. i have never worked with subds before, but it seems to me that my workflow is pretty standard. i guess if i converted to subds without any additional modeling, this could have been avoided, but what's wrong with continuing to model at the subd level?

Well actually its quite uncommon for people to use the subd refinement levels. Because in the past renderers dindt support these hierarchy edits. indeed they stll dont for most part, tough prman does.

Its not a question of not being able to send in polygons, but its just not smart since you loose the whole range of benefits that the surface that actually has a limit supplies to the render. And a lot of perofmance.

Third, its not always optimal to make sub level. Becase sometimes actually modeling the thing again makes for a subdiv thats more optimal and thus faster and thus renders faster because it dices faster.

fourth, maya has 4 different subdivision schemes, so your just choosing one of them.

[goldrake]

QUOTE
Well actually its quite uncommon for people to use the subd refinement levels.

but i see people using and recommending this technique all the time... here's ONE EXAMPLE, and there are many others. I mean, how else would one work in subds?

QUOTE
Because in the past renderers dindt support these hierarchy edits. indeed they stll dont for most part, tough prman does.

maya software renderer also supports the refinements (maybe not as efficiently, but they show up in the renders).

Not sure I understand these two paragraphs...
QUOTE
Its not a question of not being able to send in polygons, but its just not smart since you loose the whole range of benefits that the surface that actually has a limit supplies to the render. And a lot of perofmance.

Third, its not always optimal to make sub level. Becase sometimes actually modeling the thing again makes for a subdiv thats more optimal and thus faster and thus renders faster because it dices faster.