PDA

View Full Version : Layers in Modeler



lwnewbie
03-29-2005, 07:29 AM
Hi Guys, and Gals,
I was told that rendering 1 layer models is faster than rendering models that have many layers i.e. 6 or more. Is this true? Also would reducing the size of the model in layout decrease rendering times? Lastly, How does grid size in layout affect render times? I'm using LW7 on a 2.4g P4. Any and all replies would be appreciated. Thanx. :)

WilliamVaughan
03-29-2005, 08:52 AM
I'm not sure why any of the items you listed would affect render times.

lwnewbie
03-29-2005, 09:19 AM
Thanx for the reply proton.
This is what I'm trying to find out. Maybe it was a stupid question. I will have to do test renders and see for myself. Thx :o

Matt
03-29-2005, 11:30 AM
Not really ...

Setting the Grid Size just allows you to navigate around your scene in Layout faster or slower - depending on your scene.

Making an object smaller does just that, it's still the same object. Although if making it a lot smaller meant it was futher away from the camera, then that would have an impact. Closer to camera = more detail = longer render times.

Number of layers I don't think has much impact, maybe if you had hundreds of layers there might be a small increase in render time as layers in objects are treated kind of like seperate objects, and LightWave needs to work out what it needs to render, in which order, which lights are hitting etc.

hrgiger
03-29-2005, 12:29 PM
I'm not sure about the Lightwave renderer but I've found that breaking things up into layers lets fprime render faster.

brap
03-29-2005, 01:41 PM
Speaking of grid scalling, I've always wished there was was an automatic scalling function for the pan/zoom tools, kind of how it works in Modeler and some other apps: Auto-fit to a selection, and have the pan/zoom automatically scale to the proper ratio.
Anyone know if there is some way to do this, or if it's already in the feature request list somewhere?

lwnewbie
03-30-2005, 09:03 AM
After doing some test renders, I did find that layers in modeler doesn't matter, and resizing the model in layout does cut rendering time by about 1/3, depending on the closeness of the camera, and of course the model itself. I also found that adding a transparent rectangle in front of the camera reduces render times. I got that tip from a thread in this forum, but forgot where the thread is. Thanx guys for ur help. :)

lwnewbie
03-30-2005, 09:08 AM
Brap, I was at ur site. That camera looks FANTASTIC!!!!!! Great Renders. I hope to get results like that one day.

Silkrooster
03-30-2005, 07:06 PM
The object your scaling, does it happen to be sub patched?
Silk

brap
03-31-2005, 06:41 AM
lwnewbie, thanks, that's very kind of you. Those are FPrime renders, which really cranks though radiosity renders for technical stuff like that very quickly.
-Jamie

Lukesutherland
03-31-2005, 11:18 AM
I also found that adding a transparent rectangle in front of the camera reduces render times. I got that tip from a thread in this forum, but forgot where the thread is. Thanx guys for ur help. :)


can anyone else verify this?? how weird...

so do I parent a single polygon with 100% transerancy to the camera and my renders will be 1/3 quicker? WoW! Definately gonna try that!

hrgiger
03-31-2005, 12:38 PM
It only works in some situations.

Another rendering tip: If you have objects sitting on a 'floor' object(or any large flat surface), subdivide the floor into smaller polygons. If the floor renders first, it's possible that a lot of surface will be rendered that will never be seen anyway and subdividing the object can change the order that things are rendererd so it will most likely save you some time.

Para
03-31-2005, 02:37 PM
can anyone else verify this?? how weird...

so do I parent a single polygon with 100% transerancy to the camera and my renders will be 1/3 quicker? WoW! Definately gonna try that!

It's degrades image quality so it's not really worth it (unless you're willing to sacrifice image quality of course). I'll try to explain it the way I remember it is - the explanation may not be 100% correct but it's close.

What it does is that basically the first bounce (rays hit the transparent surface) is rendered at full resolution and then the second bounce (rays hit the surfaces behind transparent poly) are calculated at half resolution. This is standard way for LW to do reflections and refractions since in normal use that doesn't cause artifacts or anything like that to render, at least not to my knowledge. So, with this transparent polygon trick the renderer is fooled to render things at half resolution and interpolate big portion of the scene itself which of course shows speed increase but causes some loss of image quality.

Lukesutherland
03-31-2005, 02:50 PM
Thanks Para, I appreciatte the explanation.

So the gain would only be there if you had either raytrace transperancy, reflections or refractions turned on as a render option? If you had none of those there would be no gain?

Para
04-01-2005, 12:49 AM
I don't know actually, I'd have to run a batch of tests to see if it helps in any of the cases you just mentioned but I won't since that trick isn't interesting enough for me :P

Lukesutherland
04-01-2005, 03:17 AM
fair enough, might have a play around this weekend with it - I have a big ole scene to try this out on

cheers

Matt
04-01-2005, 11:07 AM
Hey Brap!

That Sony camcorder on your site is awesome!!! Any chance of seeing a wireframe of that anywhere?

brap
04-01-2005, 01:06 PM
Yes, will definitely have to try that poly in the frame trick!

Matt, thanks! Yes, I posted that a ways back during it's development. There's a wire pic in there.
http://www.cgtalk.com/showthread.php?t=130533

cheers,
Jamie

Nemoid
04-02-2005, 08:51 AM
if u have subpatched objects in your scene wich don't deform, freezing them renders fast :D
Lw loves poligons !

tischbein3
04-02-2005, 10:30 AM
well my two cents:

seems spliting up the object in multible layers / objects could make render time smaller, under certain conditions (see about the "raytrace otimization option" in the manual).(I will verify this in the next month), but in general, I think, the work-time you have to invest is far greater than the speed gain.

Rendering transparency trick: Gives some speedup if the polygon depth order is disrupted like mentioned above by hrgiger.

Resizing objects and camera:
I think this could rather result in more rendering errors, because of floating point number roundings... especially if you use radiosity.
but this is just a guess, and wasn't verified by me.

tischbein3
04-02-2005, 09:24 PM
ok, finished some testing.. (now is allways better than later)

I decided to use a bunch of leaves of a tree object, because I could check several issues:
Large amount of simple ( 18 faces) polygonal objects,
with transparency and some cross intersection.


First I've split up the object in single object leaves, (1494) and created two scenes, with exactly the same settings: One area light and camera. (position of them in both scenes the same)

So here is the result (rendering optimisation turned off in brackets):


leaves as a single object with transparency channel:

no raytrace shaddow, no raytrace transparency: 12,1 s [12,1s]
raytrace shaddow, no raytrace transparency: 7,11m [7,31m]
raytrace shaddow, raytrace transparency: 7,8m [7,27m]

leaves as a single object without transparency channel:

no raytrace shaddow, no raytrace transparency: 1,4 s [1,4s]
raytrace shaddow, no raytrace transparency: 4,1s [4,4s]

leaves splittet in multible objects without transparency channel:

no raytrace shaddow, no raytrace transparency: 9,3 s [9,2s]
raytrace shaddow, no raytrace transparency: 26,43m [26,59m]
raytrace shaddow, raytrace transparency: 27,27m [27,44m]

leaves splittet in multible objects without transparency channel:

no raytrace shaddow, no raytrace transparency: 1,7 s [1,7s]
raytrace shaddow, no raytrace transparency: 1,19m [1,19m]


(Previously I did the tests with a distant light, but because some other applications were also running in background, I didn't trustet the results, and did it again)

Morale:

My expectation was completly in the opposite direction: I thought the single object would slightly outrun the multible in the no raytrace shaddow pass, because LW has to handle more data, but quickly fall back in the other tests.
So I'm a little bit surprised by the result. (Especially how big the difference is)


Maybe this result will change if you use radiosity or multible lights, or these leaves are not double sided or intersect with others (etc)....or objects with more volume... I don't know.

(I also did some additional tests by centering the pivot of each object to the center of the mesh, (one time by moving the pivot position and one by centering the object (and pivot) at 0,0,0 and moving the object in layout), suspecting a problem in calculating the bounding volume,
but without any significant changes in render time)

Also a big "no" to the multible objects scene was the enormious loading time: a few seconds in comparison to 330 seconds. (Not to mention the working time of the script(s) I've used to split it up.)


So maybe I've done something really wrong or I have to agree to the first post.