PDA

View Full Version : AntiAliasing tests


Chart
02-05-2003, 01:50 AM
We all know how bad the antialiasing is in version 10.
A thread started on the list about how to work around
it. Here is Steve's reply to the thread after about 3
responses:

---------------------------------------------------

This is simply astounding! We are aware of NO aa problems
in v10! if you'd care to provide us with very exact examples
and data, we'd be happy to look at it!
This thread should NOT be continued on the animaster list!

Steve Sappington

---------------------------------------------------

Just when you thought they may have learned. Well, send
examples to Steve if you've got them...

Chart

koon69
02-05-2003, 03:32 AM
I think this is not a bad thing at all. I have read so many people say that AM sucks with AA - me included. Well it looks like Steve just opened the door for people to send their stuff in. Either we - and me - shut up or send files in. Me Im sending my files ASAP! :-) I think it would be cool if we post the images here to. And see what kind of response we get from Steve and Hash. I think Wegg and Ballistic would be the best guys in getting this shown to Steve and Hash. Those guys rule and they know Hash like crazy. Im betting that they can whip up some crazy s@$t to show Steve and Hash!!!!

Raist3d
02-05-2003, 10:34 AM
Originally posted by koon69
I think this is not a bad thing at all. I have read so many people say that AM sucks with AA - me included. Well it looks like Steve just opened the door for people to send their stuff in. Either we - and me - shut up or send files in. Me Im sending my files ASAP! :-) I think it would be cool if we post the images here to. And see what kind of response we get from Steve and Hash. I think Wegg and Ballistic would be the best guys in getting this shown to Steve and Hash. Those guys rule and they know Hash like crazy. Im betting that they can whip up some crazy s@$t to show Steve and Hash!!!!

I think Chart's point is that lack of better AA is so obvious that it should be a "ok, we are aware of it, and working on it or doing this about it" kind of deal. It's been like this for a while, this is not new.

On the other hand, how to refuse Steve's offer? Benefit of the doubt given, send files to him (i.e. I agree)

- Raist

runejw
02-05-2003, 11:59 AM
The best way to illustrate AA problems is to post an example (or two).

Perceptions differ, and what is a detail to some, can be significant to others.

When working with AM 8.5 I found the best AA-effect by using the oversampling option (not the other AA options). Made cleaner renderings in my opinion. The downside is that oversampling requires more time and huge amounts of RAM.


Cheers,
Rune

Chart
02-05-2003, 01:55 PM
Someone on the Hash list posted example images comparing the new and old render's output. I don't have the email anymore but it was a good effort to document the differences. I am sure they are no longer on the Hash list. :rolleyes: Does anyone know who posted these images?

Chart

zandoria
02-05-2003, 04:19 PM
If you have a problem with anti-aliasing in V10, send it in. Since Hash made changes to the renderer in V10, it would be a waste of everyone's time to send in an image created in any earlier version...:shame:

Wegg
02-05-2003, 05:15 PM
I'd be very interested in seeing these.

I did a quick test just now exploring the different AA choices in Lightwave.

http://www.3d-geeks.com/Wegg/NoAA.gif

You can see how I set the shot up here. .

http://www.3d-geeks.com/Wegg/LW.jpg

And you can download the object I used here. . .

Object (http://www.3d-geeks.com/Wegg/AATest.zip)

I don't think the issue is wether or not AA in Animation Master is "Broken". I just think they have taken away any choice an artist has in how to deal with the Jaggies. From what I remember. With AA turned on. . . and at 200% oversample. . . you get about the same qality as Lightwave's "Low" setting. Never mind the fact that your render times have shot through the roof at those settings. . . :thumbsdow

PJC
02-05-2003, 05:41 PM
Okay,

I set up similar projects in AMv10 and C4Dv8 to show everyone what the Eggman is talking about.

Download at your pleasure and go frame by frame.

It's not as bad as I thought it would, but it could be better in AM.

AM version of checkertest [Quicktime] (http://www.ccpots.com/PJCdemo/AMTESTS/checkersAM.mov)

Cinema4D version of checkertest [Quicktime] (http://www.ccpots.com/PJCdemo/AMTESTS/checkersc4d.mov)

- pjc

Wegg
02-05-2003, 05:53 PM
And another quick test. . .

This is just a checkerboard with the camera turning slowly.

If ANYONE can re-do this in the latest version of AM using their best AA settings. . . that'd be cool.
http://www.3d-geeks.com/Wegg/Checker.gif

pdaley
02-05-2003, 05:55 PM
If you really want to throw down, replace one of the checkers (black or white) with a well-defined texture like a marble. Then you'd get to see the shimmering, too.

koon69
02-05-2003, 06:01 PM
Not to bad but like you said can be better. I think your links should be sent to Steve ASAP! Love to hear what he says.

Wegg
02-05-2003, 06:13 PM
PJC those were GREAT tests. I can really see the difference even with the camera up that high.

If you could do one with just black and white and right down near the checkerboard. . . that'd really help us compare the three renderers. C4D, AM and LW.

Great job dude. . . :thumbsup:

Chart
02-05-2003, 06:45 PM
Wegg,

Thanks for getting the tests started! :thumbsup:

Could you post your checker test in a mov format? I don't think the limited palette of a gif doesn't show the true quality of LW does it?

Chart

My Fault
02-05-2003, 06:47 PM
Awesome job there guys! It's nice to see actual examples of the issues.

Wegg
02-05-2003, 06:55 PM
Originally posted by Chart
Could you post your checker test in a mov format? I think the limited palette of a gif doesn't show the true quality of LW.

Chart

Thats actually what I LIKE about .gif

You only get 256 levels of grey. . . so your really isolating the issue.

With .mov. . . you never know what kind of "optomizing"(read blurring) your gunna get on your image.

But. . . here ya go anyway

Movie (http://www.3d-geeks.com/Wegg/AA.MOV)

Chart
02-05-2003, 07:18 PM
Wegg,

You are right about gray scale GIFs, I was thinking color.

The reason I was asking is the LW tests don't look as good as I thought they would. What AA level are you using?

Chart

Wegg
02-05-2003, 07:25 PM
High.

This is the harshest test you can give any software.

black on white harsh lines at almost every angle. . . MOVING.

No software is perfect. In-fact if you were to go out onto a giant black and white checkerboard with a really high res camera and film. . . the results would look "jaggie" if you digitized the results and made them into a quicktime.

Its just the nature of pixels.

I'm just trying to establish the fact that some/all software does this better than AM.

PJC
02-05-2003, 11:29 PM
Here's the latest:

Amv10
Time: 5:46
320 X 240
Sorenson Compression, Best, 30fps, keyframe every 10.
AA turned on, Temporal Filter, Dither, Softness .1
AMv10 Checker test 2 (http://www.ccpots.com/PJCdemo/AMTESTS/checkers2AM.mov)

C4Dv8
Time: 2:40
320 X 240
Sorenson Compression, Best, 30fps, keyframe every 10.
AA turned on. Best AA, Sinc Filter set at 1, 4 (default for Sinc)
C4Dv8 Checkertest 2 (http://www.ccpots.com/PJCdemo/AMTESTS/checkers2c4d.mov)

As you can see, AM isn't quite as nice, but C4D is about the best out of box renderer I've seen.

- pjc

gra4mac
02-05-2003, 11:41 PM
I thought I'd follow along and learn something. I'm doing my comparison between v10b6 and Strata Studio Pro. I love texturing in Strata because you can manipulate the decal in real time, ie. scale, postion, rotation. It's great fun to watch the decal move around on the object.

Only one problem, in v 10how do you make tiles repeat. I can't find the settings. Help please.

Cheers, Graham

Chart
02-05-2003, 11:50 PM
Graham,

In AM I would create a material, change it's type to combiner/checker. Set one of the attributes to black and the other to white. Make the size more like 180cm. Apply this material to the default ground object.

Chart

Wegg
02-06-2003, 12:12 AM
As you can see, AM isn't quite as nice, but C4D is about the best out of box renderer I've seen.

OOoo very very nice. Perfect. I love how Cinema almost has a Depth of feild look to it.

I really hope Hash takes notice. . .

Wegg
02-06-2003, 12:17 AM
Here is. . . onikaze's checkerboard AA test from AM.

Much the same but its nice to see the shimmers come from multiple sources.

Shimmerama (http://www.3d-geeks.com/Wegg/checkertest.mov)

Its those damn AA (or lack of AA) shimmers that we are all frustrated with. And it is something that other software companies have solved. So I would consider it a "Bug".

Nonproductive
02-06-2003, 12:20 AM
/me channels Steve
"Those are not the shimmers you're looking for"

:p

Chart
02-06-2003, 01:38 AM
PJC,

The C4Dv8 AA test looks great. How long did it take to render compared to AM?

Chart

Wegg
02-06-2003, 01:52 AM
What little I know about C4D. . . probably a fraction of the time. Especially on a dual system.

PJC
02-06-2003, 04:41 AM
2m:40s for C4D v8

5m:46s for AMv10

both were rendered on my single PIII 500MHz laptop.

I'll have to check my C4D manual (it's at work), but the way it handles AA is awesome. Off the top of my head, I believe it does it this way:

Any scanline part of the render gets 16 X 16 AA, and any raytrace element is where the "filter" kicks in. In my case it was Sinc and it ranged between 1X1 to 4X4. So it wasn't even set that high for the test.

Wegg, I believe its the SAT filter falloff on the texture that's making it the DOF sort of feature.

C4D lets you set up AA for each texture, and then the geometry itself can have its own AA. Pretty powerful.

- pjc

jdates
02-06-2003, 05:07 AM
Are these single pass AA attempts?

Or Multipass?

If they are Multi-pass.. that indeed is lame.

jdates

JoeW
02-06-2003, 06:08 AM
I spent the last 3 days working (re-creating) some of the render tests I did a long time ago. I created a very simple scene and did tests in AM10, C4D and LW. I can tell you that AM's EDGE aa isn't *horrible* but it's treatment of bump maps and texture maps *is* hideous.

SO I was ready to post all of these 40-odd tests for the world to see, when into my office walks Jeff Bunker. He told me that Martin has implemented a new renderer into 10.5 that will have *perfect* AA, perfect motion blur, etc - Maritn claims he's basically put the LW renderer into AM. Martin said it's new from scratch and is "flawless."

I wanted to try it out and see for myself - but I couldnt' get it to work (what a suprise).You can download the Alpha - get the DOC file, too - it tells you how to access the new renderer.

Anyway, I thought you all would want to know. I'm looking forward to seeing this flawless renderer...

BTW - One thing that Martin said to Jeff - "I know my customers, and my customers won't tolerate a slower renderer - even if it IS perfect." I begged to differ - slow and beautiful, I can handle - slow and shitty looking is another story. If you are a "user" and you don't agree with Martin (in that you don't MIND slow if it's gorgeous) you might drop him a note (martin@hash.com).

Now, to get him to fix the surfaces....

JoeW

My Fault
02-06-2003, 06:32 AM
Wow Joe, this is pretty interesting news. It will be interesting to see what Martin has come up with. Man I hope my dongle gets here soon!

Hookflash
02-06-2003, 06:33 AM
JoeW: Please keep us posted on the 10.5 renderer, if you make any progress with it! :buttrock:

ragtag
02-06-2003, 10:40 AM
JoeW could you please post those 40 tests. Even though there is a better renderer on the way, it would be interesting to see.

Ragnar

hoochoochoochoo
02-06-2003, 11:23 AM
just to add to PJC's c4d tests, the renderer is blindingly fast AND you do get a lot of control over AA but you have to live with the limitations to modeling and animation.:shrug:

FWIW - I've only ever known cinema4D crash on me once in 4 years.:thumbsup: You could chuck a proverbial brick s***house at it and it'll still be standing looking back at you.
The version I have - version5SEat home, given away free 6 by Maxon months ago and XL, Go bought by my workplace.

This is one case where fast AND beautiful worked together. I just wish AM would try and compete here also.

Natess44
02-06-2003, 03:34 PM
Originally posted by JoeW

I wanted to try it out and see for myself - but I couldnt' get it to work (what a suprise).You can download the Alpha - get the DOC file, too - it tells you how to access the new renderer.
JoeW
So where is the alpha for 10.5 on the Hash site?

amsmf
02-06-2003, 03:51 PM
10.5 alpha is on the ftp...

f t p://ftp.hash.com/pub/updates/windows/Am2003/alpha/

Now... if someone can just get it to work and post some tests.

Wegg
02-06-2003, 03:56 PM
BTW - One thing that Martin said to Jeff - "I know my customers, and my customers won't tolerate a slower renderer - even if it IS perfect." I begged to differ - slow and beautiful, I can handle - slow and shitty looking is another story. If you are a "user" and you don't agree with Martin (in that you don't MIND slow if it's gorgeous) you might drop him a note (martin@hash.com).


VICTORY! Everyone pat yourselves on the back. I don't know what chain of events caused Martin to realize the renderer needed an overhall. . . but I'm sure our little rebellion didn't hurt. Now we just have to wait and see. . .

Natess44
02-06-2003, 04:01 PM
Originally posted by Wegg
VICTORY! Everyone pat yourselves on the back. I don't know what chain of events caused Martin to realize the renderer needed an overhall. . . but I'm sure our little rebellion didn't hurt. Now we just have to wait and see. . .
You should see what he wrote about it. It's on the ftp site, sounds much better.

koon69
02-06-2003, 04:07 PM
I couldnt log in to the site? Any idea how to get in and download? I tried as anonymous but the system froze.

koon69
02-06-2003, 04:09 PM
Got in. Read the doc sounds great. If I have time Ill have to give it shot. Any idea when 10.5 is out?

Wegg
02-06-2003, 04:29 PM
Is it too long to post here?

PJC
02-06-2003, 04:29 PM
Originally posted by hoochoochoochoo
just to add to PJC's c4d tests, the renderer is blindingly fast AND you do get a lot of control over AA but you have to live with the limitations to modeling and animation.:shrug:


Version 8 has improved every facet of its animation toolset. It's character stuff is good (better than stock LW, but not AM or Messiah) and the modeling tools are WAY better than the old 5 days.

I still like AM tho ;) and with a new renderer, YIKES!

- pjc

amsmf
02-06-2003, 04:31 PM
the 10.5 doc:

VERSION 10.5 Alpha

"Programming effort at Hash is divided into three categories: refinement of Version 10 Release (50%), Macintosh OSX port (40%), and research (10%). As is our strategy, 10.5 Alpha contains changes deemed too risky (stability wise) to put into10 Release. (Simple changes are put into both code bases).

In 10.5 Alpha the renderer has been changed so that with “Antialias” “Off”, a simple point-sampled, z-buffer renderer is used. The point-sampler can also use “multipass”, which renders each frame multiple times (for antialiasing), with slightly different camera positions (for depth of field), at incremental time steps (for motion blur), and the lights are jittered (for area lighting and penumbral shadows). However, it will require many passes (64) to achieve the same edge softness as “Antialias” which uses a more efficient a-buffer, so “Antialias” “On” will be much faster (6 to 20 times)."

That's the entire doc. I didn't see anything in there on how to access the renderer, as I believe JoeW suggested.

Natess44
02-06-2003, 04:31 PM
Originally posted by Wegg
Is it too long to post here?
"VERSION 10.5 Alpha

Programming effort at Hash is divided into three categories: refinement of Version 10 Release (50%), Macintosh OSX port (40%), and research (10%). As is our strategy, 10.5 Alpha contains changes deemed too risky (stability wise) to put into10 Release. (Simple changes are put into both code bases).

In 10.5 Alpha the renderer has been changed so that with “Antialias” “Off”, a simple point-sampled, z-buffer renderer is used. The point-sampler can also use “multipass”, which renders each frame multiple times (for antialiasing), with slightly different camera positions (for depth of field), at incremental time steps (for motion blur), and the lights are jittered (for area lighting and penumbral shadows). However, it will require many passes (64) to achieve the same edge softness as “Antialias” which uses a more efficient a-buffer, so “Antialias” “On” will be much faster (6 to 20 times)."

Nonproductive
02-06-2003, 04:45 PM
I'd be very interested to see some sample renders out of 10.5 - and render times as well ;)

I just got my upgrade flyer in the mail - some highlights:
"world class renderer"
"self evident interface"
"scripting, expressions, relationships, layers"

that last one stuck out for me - since Martin basically said that 9 sucked *because* of them - yet they are on the first line of the v10 upgrade flyer...*shrug*

ypoissant
02-06-2003, 04:48 PM
I just tried it.

The way to make the multipass go into action is you go to your camera (the object, not the camera shortcut un choreography) and set "MP" ON. Then you can set the number of passes. I left mine at 9. I turned depth of field OFF and the other one OFF too (I don't remember how it was called). I think the AA option have to be turned OFF first too.

Then rendered ma favorite AA test project.

BINGO! This is exactly what I was looking for. And the render time is only 2.5 the time of AA alone.

Yves Poissant

My Fault
02-06-2003, 04:50 PM
Originally posted by ypoissant
I just tried it.

The way to make the multipass go into action is you go to your camera (the object, not the camera shortcut un choreography) and set "MP" ON. Then you can set the number of passes. I left mine at 9. I turned depth of field OFF and the other one OFF too (I don't remember how it was called). I think the AA option have to be turned OFF first too.

Then rendered ma favorite AA test project.

BINGO! This is exactly what I was looking for. And the render time is only 2.5 the time of AA alone.

Yves Poissant

Can someone throw the antialias test in to 10.5 and post results and times?

Nonproductive
02-06-2003, 04:53 PM
the render time is only 2.5 the time of AA alone
which equates to....? How much time in 10 and in 10.5? and can you post a side by side render?

I'm with JoeW on this - I can tolerate slow if the quality is there. I can always drop options to render a draft for quick timing tests but I can't add quality.

JoeW
02-06-2003, 05:02 PM
Just FYI - either they've hidden it really well, or it's not there - but I've been unable to find Multi-pass rendering in Netrender this version.

SO - If you're thinking about buying Netrender so that you have that capability, you might want to ask Hash where it went.... and while you're at it, ask what the hell happened to "Rays Cast" for raytraced shadows...

JoeW

Natess44
02-06-2003, 05:06 PM
v10 net doesn't have the multipass but alpha 10.5 does and so does the normal version.

ypoissant
02-06-2003, 05:34 PM
Joe,

I have and did use the Net version of v10.5. and the multipass is there. Maybe you have to turn the advanced properties ON in the options. Then, as I wrote earlier, there is a MultiPass option in the camera object under the output drop-down.

I am preparing some images and rendertimes right now. But it is possible that there is more than what I already found. It seems that it is possible to turn off a-buffers entirely and then save even more time at rendering. I'll post

Yves

Wegg
02-06-2003, 05:40 PM
OOoo that sounds cool.

Keep us informed!

ypoissant
02-06-2003, 06:20 PM
OK here goes the data:
The test project was the "Toys" project with all bitmaps removed except for bumpmaps. I wanted to have a monochrome image in order to clearly see the aliases. The original render was 600 x 320. The full scene is there.

What I show here is a portion of the render enlarged 3x. There are two areas that I use to evaluate the AA efficiency: The top edge of the inset on the bloc and the shadow on the ground. The top edge is tough because the shading is part geometry, part shadow.

http://www.ypoart.com/show/2o-v10.5-NA.jpg
This is non-antialiased and no multi-pass.
Render time: 0:00:35

http://www.ypoart.com/show/2o-v10.5-AA.jpg
This is with AA turned ON.
Render time: 0:01:55

http://www.ypoart.com/show/2o-v10.5-9MP.jpg
This is with AA turned OFF and 9 samples multipass.
Render time: 0:04:56

http://www.ypoart.com/show/2o-v10.5-16MP.jpg
This is with AA turned OFF and 16 samples multipass.
Render time: 0:08:46

As for this extra feature of disabling a-buffer completely, it seems that this is it.

Yves Poissant

Chart
02-06-2003, 06:25 PM
I emailed Steve the 2 PJC render test links. Here is is reply...

----------------------------------------------------------------

I would suspect the Cinema4D output has been blurred- if
you apply the same in A:M (Soften, in the post section of the
render setting) it will look similar- In any case, DO NOT send
these examples unless you send data as well-

----------------------------------------------------------------

So should PJC send project files to Steve or should we just wait for 10.5?

My Fault
02-06-2003, 06:28 PM
Originally posted by Chart
So should PJC send project files to Steve or should we just wait for 10.5?

When in doubt, send data!

Nonproductive
02-06-2003, 07:25 PM
I'm going to refrain from commenting on the 10.5 renderer until we can see more stills and animation...shadows, etc...

However - In Chart's post - is Steve implying that A:M's render in v10 is comparable to C4D's if you blur A:M's output?!? Please...someone with the AA test scene - do the test with Steve's suggestion...I find it hard to believe.

Shame on you PJC for cheating and blurring the C4D render to hide its poor AA!

PJC
02-06-2003, 07:27 PM
Guys,

I talk to Steve all the time, I'll get him data. All I ask is that you don't send him stuff from me. I'll take care of it, okay? :buttrock:

My AM example DID have soften turned on, setting was .1.

C4D has the advantage because you can AA the texture AND AA the scene, but what Steve might be talking about is the MIP strength in C4D, which might be like soften in AM. I'll post soon.


- pjc

Chart
02-06-2003, 07:36 PM
PJC,

Sorry, I should have asked you first.
I talk to Steve all the time, I'll get him data. All I ask is that you don't send him stuff from me. I'll take care of it, okay?
Chart

PJC
02-06-2003, 07:39 PM
that's cool. :beer:

- pjc

PJC
02-06-2003, 08:38 PM
Okay, here's the skinny on C4D.

I believe it does three type of AA.

It has a setting for AA on geometry. All sorts of options: "Sinc, Area, Cone, Catmull, etc."

It has a setting for AA for Materials (bitmaps and algorithmic) Same options for those as gemetry.

And there is a final AA for the scene. Same options as above. I believe this is where AM's "Soften" filter would be utlized.

I believe (I could be wrong) that the top two are done as it's rendering, and the third "scene AA" is done post image kinda like AM.

So, there you go.

I agree that I would love to have OPTIONS! Not just AA on or off. I would love to control the AA algorithm: Some are better for still images and some are better for film, and others for video.

- pjc

Chart
02-06-2003, 08:53 PM
Yves,

Thanks for posting the v10.5 images. The 16 samples multipass does look better but at almost 4.5 TIMES the render time of the AA render. The time is about 16 times the non-AA render so I guess it really was rendered 16 times. It seems that this is a brute force approach. With so many passes over the image, it seems there has got to be a better way.

Speaking of brute force, could you render the same image with say 200% oversampling? It would be interesting to compare render times and images.

Chart

ypoissant
02-06-2003, 09:10 PM
Chart,

200% oversampling would be equivalent to 4 passes. 9 passes equivalent would be 300% oversampling, 16 passes = 400% oversampling.

I did those renders yesterday with v10 but did not record the render times. I think it was 26 minutes. BTW, I could render a 2400 x 1280 (a 400% oversampling) yesterday without a hickup from 10.0. I remember it took about 45 minutes to render though. And I only have a p-III 800mHz with 128Meg RAM.

I shall do the Oversampling renders again and post the results. But that will be tomorrow.

BTW: About C4D AAs, Those "Sinc", "Area", etc, are signal filter functions. They can be applied at render time and as post processing, The result is very different of course. It can also be applied to texture. A:M also does apply filters on bitmap textures. I believe they use bi-linear which would be equivalent to a box function. They also use a filter function at render time but I don't know which one. And there is the blur post processing which should be bi-cubic which is roughly equivalent to sinc or gaussian.

Yves Poissant

JoeW
02-06-2003, 09:29 PM
Okay, my bad - I asked people to e-mail Martin, and I probably shouldn't have.

Right now there is some VERY positive momentum going with the renderer. I can't say exactly what - I've been asked not to - but I can assure you all that it IS good.

I'm asking that folks (for the time being) not deluge Martin with e-mails concerning the renderer. I know he's working very hard on it right now, so let's all hold our tongues until we see what he comes up with - you never know, we *may* get what we wish for.......

Sorry if I led anyone astray.... :)

JoeW

PJC
02-06-2003, 09:42 PM
Originally posted by ypoissant

BTW: About C4D AAs, Those "Sinc", "Area", etc, are signal filter functions. They can be applied at render time and as post processing, The result is very different of course. It can also be applied to texture. A:M also does apply filters on bitmap textures. I believe they use bi-linear which would be equivalent to a box function. They also use a filter function at render time but I don't know which one. And there is the blur post processing which should be bi-cubic which is roughly equivalent to sinc or gaussian.

Yves Poissant


Awesome information Yves! Thanks!

- pjc

Chart
02-06-2003, 10:31 PM
All this new render stuff is great. Now if they will support multiprocessors so we can put together killer machines.;) It's a new toy to play with.

Right now there is some VERY positive momentum going with the renderer. I can't say exactly what - I've been asked not to - but I can assure you all that it IS good.

Chart

My Fault
02-06-2003, 11:09 PM
Originally posted by JoeW
Okay, my bad - I asked people to e-mail Martin, and I probably shouldn't have.

Right now there is some VERY positive momentum going with the renderer. I can't say exactly what - I've been asked not to - but I can assure you all that it IS good.

I'm asking that folks (for the time being) not deluge Martin with e-mails concerning the renderer. I know he's working very hard on it right now, so let's all hold our tongues until we see what he comes up with - you never know, we *may* get what we wish for.......

Sorry if I led anyone astray.... :)

JoeW

Yeah, I just got off the phone with Steve and I am pretty psyched! Props to everyone who sent in detailed render issues with data and HUGE thanks to JoeW and the rest of the Avalanche crew for all their work...awesome!! :thumbsup:

ypoissant
02-06-2003, 11:10 PM
Here are the facts so far:

This a-buffer disabling I mentionned is really there. Actually a-buffer only enter in action when you explicitly set antialias ON. So the rendering times are way much more faster than in v10.0. For instance, the same scene rendered with the same specs gives those timings:

System: P-III, 800mHz, 128Mb

Antialias OFF:
V10.0: 0:01:09 (a-buffer in action)
V10.5: 0:00:35 (a-buffer disabled)

Antialias ON:
V10.0: 0:01:49
V10.5: 0:01:55

Antialias OFF, 200% Oversampling (or 4 passes):
V10.0: 0:05:19
V10.5: no data yet.

Antialias OFF, 300% Oversampling (or 9 passes):
V10.0: 0:20:59 (0:11:00?)
V10.5: 0:04:56

Antialias OFF, 400% Oversampling (or 16 passes):
V10.0: 0:42:52 (0:19:00?)
V10.5: 0:04:56

Because of 128Mb, oversampling more than 200% caused data swaping to disk so timing is off. I included an estimate of what would probably be a more exact timing in parentheses.

Yves

Nonproductive
02-06-2003, 11:24 PM
MyFault wrote:
Yeah, I just got off the phone with Steve and I am pretty psyched!

care to share anything (if you can)? Is this 10.5 stuff or 11 stuff? Rendering issues only or stability? So far we've seen AA is being addressed - any of the other issues mentioned here?

Was this "THE Discussion" with Steve or just shooting the breeze? If it was, in fact, the community discussion - anything to bring back?

My Fault
02-06-2003, 11:30 PM
Originally posted by Nonproductive
MyFault wrote:
care to share anything (if you can)? Is this 10.5 stuff or 11 stuff? Rendering issues only or stability? So far we've seen AA is being addressed - any of the other issues mentioned here?

Was this "THE Discussion" with Steve or just shooting the breeze? If it was, in fact, the community discussion - anything to bring back?

It was a bit of both and I'm typing something up that I'll post in a separate thread. Can't talk about everything, but I am definitely pretty psyched!

Nonproductive
02-06-2003, 11:33 PM
excellent Thanks! :curious:

ypoissant
02-07-2003, 01:31 PM
Just to mention that I updated the data in my previous post up here.

That is strange because I did exactly that yesterday night and this morning, everything in this thread reverted back to what it was before I posted. What happened? Server crash?

Yves

Chart
02-07-2003, 02:10 PM
Yves,

Thanks for mentioning to scroll up for your message. I was looking for it but expected it to be at the end.

Version 10.5 multi-pass appears to be way faster then over-sampling. Could you post the block images comparing over-sampling and multi-pass?

Now that output scaling has been removed, the over-sampled image’s quality also depends on how the image is scaled.

Thanks for your effort!

Chart

jayrtfm
02-07-2003, 03:19 PM
Originally posted by ypoissant

That is strange because I did exactly that yesterday night and this morning, everything in this thread reverted back to what it was before I posted. What happened? Server crash?

Yves

yup, sort of, see: http://www.cgtalk.com/showthread.php?s=&threadid=42435

Kentaurus
02-09-2003, 06:41 PM
Are we talking about AA in general in this thread? In that case I might have some input. I have made a test where I compared texture AA with geometry AA and it came out as can be seen in my attachment. The left image shows a plane with a black and white checker texture applied on it. The right one shows a grid were the individual patches are colored black and white.

As I see it, from this test I can draw two conlusions about the renderer in v10.0e:
There is no textures AA at all
The geometry AA is rather good


K.T.

PJC
02-09-2003, 07:25 PM
Cool!

I was thinking of doing that test too! AM has always handled texture maps differently than any other program, and AA with those textures has always been kinda wierd in my opinion.

Good test and example of the difference!

- pjc

Wegg
02-10-2003, 03:39 PM
Very good deductive skills Kentaurus.

Natess44
02-10-2003, 04:18 PM
So maybe we should beg Hash for aa on textures too.

Wegg
02-10-2003, 04:44 PM
I'll bet Hash doesn't AA textures as a way of accelerating the renderer. Its one of those things that only "Pro" users would ask for.

ypoissant
02-10-2003, 04:55 PM
Texture is a very generic term. What you are refering to are materials or procedural textures.

A:M does AA texture maps through mip-mapping.

But it is very difficult to antialias some types of procedural textures in any other ways but through sub-sampling. Checker is such textures.

Unlike noise based procedural textures which given a pixel size in texture space, the algorithm can decide when to stop fractalize the details of the texture. This way, the procedure can make sure the noise frequency will never exceed the screen resolution resulting in automatic AA.

There is no such fractalisation in textures like Checkers or Sphericals which, because of the extremely sharp transition between black and white part of the texture, will automatically produce extremely high frequencies. The procedural have no way of determining what pixel proportion the shade will occupy and thus is incapable of computing the proper AA value for each pixel. Even with the pixel size, the procedural would need to know the surface in order to compute the intersection with the texture. The only way would be to basicaly implement a mini renderer inside the procedural as well. This would a great loss of processing power.

One workaround is to raise the "blur" value of the checker to some eye-comfortable value. Bluring the checker will remove the high frequencies from the texture and you will get more acceptable AA results on those type of procedurals.

So the easiest way to AA procedural such as checker is through sub-sampliing which is a brute-force approach. But this is the approach that LW and C4D have decided to take it seems. This is also what the v10.5 does with the multi-pass renderer.

Yves

zandoria
02-10-2003, 05:13 PM
I would rather not have A:M Antialias my textures... I like the control that I have painting alpha channels by hand, and I would rather have my textures exactly as they are painted in Photoshop.

If I want a soft edge to a texture, it seems more controllable with pixel resolution and alpha channels. That's just my preference and 2cents...

JoeW
02-10-2003, 05:57 PM
Texture AA'ing doesn't mean "soft" or "blurry" textures - it often yeilds a sharper image - instead of an edge being indistinct, it can make it very sharp and clear. I prefer that the maps on an object be representative of what I've painted, and find that AM has a way *currently* of kind of muddying over nuances in a map.

AM *currently* doesn't do a good job with maps at all. I'm not sure exactly what it's doing, but it looks like it's using a simple box filter to downsample - and that yeilds a pretty ugly result.

Here's a link to an image of a few tests I did in AM v10 and Lightwave:

ftp://ftp.hash.com/users/joewllms/samples/AA_Samples.jpg

These "swatches" are sampled right off the original renders. The texture maps used were 1186 X 289. I think it's pretty obvious that this is another area of the renderer that needs some work. I really like how the environment map on the LW render (purplish sphere) is distorted by the bump map, but remains sharp. I couldn't even come close to that in AM (currently, that is).

I hope that pretty soon I can re-do these tests and report all this is "fixed"...

JoeW

binder3d
02-11-2003, 02:54 AM
Any chance of comparing these tests with other apps? Then send to Hash to show them how far behind they are? But they might not care. Martin did say they are targeting the beginners.

zandoria
02-11-2003, 01:48 PM
I re-rendered the swinosaur leap that I posted last week, using V10.5alpha with the "multipass" option 16(4x4)

my rendering time went from 30seconds per frame to 3 minutes per frame (D1 resolution)--but there was a huge difference in quality:buttrock:

http://www.zandoria.com/images/Movies/leap.mov
3MB sorenson3

Wegg
02-11-2003, 03:26 PM
Much much nicer.

Raist3d
02-11-2003, 06:48 PM
Looks *TONS* better quality rendering wise that I have ever seen A:M do.

Nice.

JoeW
02-13-2003, 12:25 PM
Originally posted by binder3d
Any chance of comparing these tests with other apps? Then send to Hash to show them how far behind they are? But they might not care. Martin did say they are targeting the beginners.

This is happening right now. Tests out of Maya, LW, etc. Martin *is* working on the renderer. so, we're all just kind of holding our breaths to see what he comes up with. Hopefully, we'll get what we want. Simply adding multi-pass to the "standard" renderer has made a big improvement.

I think that *maybe* Martin is starting to realize that his user base would increase dramatically if only the renderer was better....

JoeW

koon69
02-13-2003, 02:01 PM
Allright Joe!!!! AM really kicks butt and now if they do get a nice render maybe people wil ltake it seriously and more people start using it. Do you think this wil lmake Hash more of a player in 3d? Or do they have to make a few more changes to be taken for real? When can we see theses test?

onikaze
02-13-2003, 05:04 PM
hmmm, thats something I would really like to see happen BUT for that to occur I think the import export would have to be more robust. This way AM could be integrated into various company pipelines.

Thats is something that probably doesn't have a high priorty but it would be nice to see it come first (after the renderer and stability) even though in Martin's response he mentions not being anyones plugin (or something to that effect).

I just don't see whats wrong with saying that you can be used with other packages though. I mean everywhere I look I see that if you get maya you NEED renederman or something else. If you go lightwave also use messiah if you want to keep sane. etc etc etc. I remeber a few years ago the mantra was model in alias and animate in softimage (or something like that)

Many houses use many different apps to get the job done and I really believe That AM could fit in there nicely for the Character animation duties. But ,for me at least, I would rather do complex mechanical stuff with a different set of tools.

hmmm I seem to be lossing my train of thought so I'll stop now :surprised :surprised

Wegg
02-13-2003, 05:17 PM
Hash Inc. has no motivation to do that because of the amount of work it would take to get something like that to happen for such a small potential market. But Raf got it to work with LW, and Avalanche have done it for other software. . . so I don't see it as impossible.. . . just not likely.

onikaze
02-13-2003, 06:11 PM
Really ?

I remeber hearing something about Raf doing that but don't remeber seeing any results.

What did he/they do ?

Wegg
02-13-2003, 06:19 PM
I know Raf was shot down pretty early on. He posted a few comments about his progress on the AM list and. . . of corse. . . had to stop talking about it. (Ah the joys of the Steve filter.).

JoeW
02-13-2003, 06:58 PM
Originally posted by koon69
Allright Joe!!!! AM really kicks butt and now if they do get a nice render maybe people wil ltake it seriously and more people start using it. Do you think this wil lmake Hash more of a player in 3d? Or do they have to make a few more changes to be taken for real? When can we see theses test?

I think it *would* get taken more seriously and used more if the renderer was up to par. I know that a lot of smaller studios would probably jump on it for use in commercials and such - I know *I* would use it more.

I think Hash will always be considered a "toy" by the uninformed and people who base their assessment of skills on how expensive the package is ("You can't be any good! AM only costs $300!"). I can't tell you how many times I've sat down with a Maya guy or a Soft guy and showed them stuff in it - only to have them sit there slack jawed - stunned with how easy it was to do such complex things. Unfortunately, AM's lack of extensibility will always hamper it with the "big shops" who like to customize the hell out of their tools - **that's** why Maya is used so much, BTW - it's not so much a 3D app as it is a *programmable 3D interface*. Every time I've seen a Maya screen in a "making of" special, I recognize parts of it, but the majority is totally custom - tools, interface, etc - definitely NOT the Maya you buy from A/W.

I've promised to not show the tests until we get a more "final" renderer solution (sorry) - and it would be pointless seeing as how the renderer is changing. I may eventually post them just to compare what the "old" renderer did compared to the new one (as well as other applications), but right now, I don't want to rock the boat as Martin *is* working on the renderer (and we should all be happy about that :)

JoeW

JBarrett
02-13-2003, 10:21 PM
Originally posted by JoeW
Unfortunately, AM's lack of extensibility will always hamper it with the "big shops" who like to customize the hell out of their tools - **that's** why Maya is used so much, BTW - it's not so much a 3D app as it is a *programmable 3D interface*. Every time I've seen a Maya screen in a "making of" special, I recognize parts of it, but the majority is totally custom - tools, interface, etc - definitely NOT the Maya you buy from A/W.

Bingo.

I've had several people write and ask me (based on the fact that I'm subscribed to the A:M list from my work address) if we use A:M here at Big Idea, and when I say that we use Maya, they almost always ask why. The key thing is Maya's extensibility (or what I call "tweakability"). Every single aspect of our production pipeline has some sort of custom tool written for it...from modeling to rigging to animating to project tracking, and on down the line. Without all those custom tools, our work would be a lot more difficult, and take a whole lot longer to produce. There's no way A:M can currently meet that need, and until a more in-depth scripting/programming capability is added (i.e. something akin to MELscript, and far simpler than coding plugins in C++), it never will.

Kevin Sanderson
02-15-2003, 12:46 AM
Originally posted by JoeW
I think it *would* get taken more seriously and used more if the renderer was up to par. I know that a lot of smaller studios would probably jump on it for use in commercials and such - I know *I* would use it more.

I think Hash will always be considered a "toy" by the uninformed and people who base their assessment of skills on how expensive the package is ("You can't be any good! AM only costs $300


I think this is possible. Just from the guys who did make a run with it like ReelFX and Eggington...they've outgrown it, but a small shop doing little projects could benefit easily. Small shops and TV stations using Carrara Studio 2 are finding it's easier to work with than Lightwave for them and they love the output. I'm sure we would see the same comments if the reported progress continues with AM.

Acceptance of CG software reminds me of what video guys tell me when I work on narrations for them. They only get respect if they have an Avid and professional BetaSP decks. They can't even mention Video Toaster or any of the other systems without being considered a "wedding" guy or similar, no matter how good their previous work has been. They experience the same frustration. DV is changing some of that but it's slow going changing perceptions.

CGTalk Moderation
01-14-2006, 08:00 AM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.