Wow, well thanks for sharing the pain. I cringe at the thought of
having to do any custom shader writing to push this job through. I
wonder if the actual amount of guide hairs effects rendering. In other
words, my viewport is almost solid with guide hairs but I'm only really
rendering hairs in areas I painted with a weight map. Do these
"ignored" hairs play any part in rendering if no render hair or instance
is applied to that area? I also wish there was a way to limit the
amount of guide hairs you see on screen. This would be much better for
interaction purposes when doing simulation.
Kris
Ben Barker wrote:
It wasn't easy for the lighters. Lots of splitting up passes. The tree
assets were designed with little thought to rendering more than one or
two, and layout ended up making thousands of them with no concern of
them rendering. We designed a proxy system and Matt Lind wrote a
custom shader that did fancy 3D shading on a 2D card and we
procedurally replaced most of the trees with those.
The renderfarm ended up being pretty awesome in the end, both hardware
and software. 64bit with gigs of memory, dedicated translation
machines, etc and that helped a great deal as well when it went
online, but there was still a good share of pain before that, and of
course you still needed real trees in many shots. At that stage
optimizing the actual tree assets was the biggest help (as opposed to
render settings which only give you so much). They had an insane
amount of texture data as well as high subdivisions and hair count, a
good deal of which could be reduced with no visible effect (SubD level
2 on 5 pixel tall parts, etc). Hair instancing, especially
recursively, multiplies any waste exponentially. Texture data was a
huge killer too, not just on trees. Being fairly brutal with reducing
this sometimes was the only way to get things to render.
On 5/31/07, Kris Rivel <krisrivel(at)gmail.com> wrote:
Thanks Ben. Sounds like you guys had the same problems I'm having. I'm
spending more time grooming and keeping turbulence to a minimum so as to
not get too much flipping. Right now I'm finding the biggest problem is
rendering. MR really doesn't seem to play nice in a case like this.
I'm optimizing to the nines but man is it slow. How did you guys manage
to get so many leaves and trees to render? Did you have a lot of bits
rendered separately? What was your memory limit on your farm? Its a
shame because I'm working on a Win64 box but I have to stick with not
only XSI 32 bit but also can't take advantage of v6 either since the
farm this is rendering on does not yet have it in place. Looks like
there's going to be a lot of coffee breaks (well I don't drink coffee so
soda or tea) for me.
Kris
Ben Barker wrote:
> I don't think it's possible to use scripted ops to control hair
> instances that specifically, since they operate on the scene data and
> hair instances don't really exist until rendering. You might be able
> to use a scripted op to scatter model instances based on guide hairs,
> essentially implement a "scatter" op, but it would be very slow
> relative to hair instancing and potentially very involved if you
> wanted it to be robust.
>
> Unfortunately if you use a global Y upvector (like the goal object you
> mentioned) instances that "aim" down that vector will still
> potentially flip during animation. There isn't really any bulletproof
> way around this as it is a problem with the philosophy of the
> algorithm not really the implementation. Careful grooming really.
>
> We had this issue on Barnyard, where all the tree leaves were
> hair-instanced clumps. This was before tangent maps and all that good
> stuff too, so careful grooming (trial and error) was the only
> solution. Pretty sure a non-trivial amount of flipping made it to
> paint and even into a few into finished shots. Huge pain in the ass :(
>
> On 5/31/07, Kris Rivel <krisrivel(at)gmail.com> wrote:
>> Is there any way to get hair instances to point up along the
global Y?
>> Ideally to have some variance as well. This possible via scripted
>> operators? I'm looking to get more rigid control over my hair
instances
>> which are incredibly difficult to keep from flipping around wildly
when
>> adding slight turbulence. I'm using tangent maps and tried
pointing the
>> instance to a goal but they don't give me the effect I'm looking for.
>>
>> Kris
>> ---
>> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
>> body:
>> unsubscribe xsi
>>
>
>
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in
body:
unsubscribe xsi
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi
