Page 4 of 29 FirstFirst ... 2345614 ... LastLast
Results 46 to 60 of 425

Thread: Radiosity, animated

  1. #46
    Quote Originally Posted by pooby View Post
    I don't know what you have in your shot, but if it has anything deforming -ie bones etc, then it is not supported by LW's interpolated radiosity. The animation cache will work only for animated 'solid' geometry
    I think that's only one of the two types of radiosity - final gather, but monte carlo works?
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  2. #47
    Quote Originally Posted by joelaff View Post
    If you still get noise then lower the threshold to 0.01. This will add another pass of AA. Though you may experiment with raising the RPE first (1 or 2 at a time), depending on whether or not you think the rest of the scene benefits from the AA, and how long each method takes.
    Adaptive of .01 will double your render times, with no appreciable results. Adaptive is not 100% efficient, which it would have to be in order to replace multiple passes of aa. 6 pass with .05 would be faster and better, things like textures will shimmer with only one pass. It's subtle, but you'd notice it if you compare the renders side-by-side. I've gone up to 12 pass with a visible improvement.
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  3. #48
    Quote Originally Posted by joelaff View Post
    With Lanczos you are really shooting yourself in the foot. That is like running an unsharp mask after each AA pass and then expecting it to remove the noise. Ever try an unsharp mask on a noisy image with no threshold? It looks like garbage.
    Totally agree. Seems that professional print people use it, but I can't understand why - it puts a white halo around sharp areas of the image - blah!
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  4. #49
    I think that's only one of the two types of radiosity - final gather, but monte carlo works?
    Neither work, but Monte Carlo will work if you dont interpolate it. However its much slower than Fprime doing the same thing.
    http://www.newtek.com/forums/image.php?type=sigpic&userid=18493&dateline=130857  4707

  5. #50
    Quote Originally Posted by toby View Post
    Adaptive of .01 will double your render times, with no appreciable results. Adaptive is not 100% efficient, which it would have to be in order to replace multiple passes of aa. 6 pass with .05 would be faster and better, things like textures will shimmer with only one pass. It's subtle, but you'd notice it if you compare the renders side-by-side. I've gone up to 12 pass with a visible improvement.
    I don't find it doubles the render time, and it definitely has an appreciable result with radiosity. Try it out. You will see significantly less noise. More samples...

    As for textures. I start with a base AA of 1. If I have texture issues I will increase it. Though I might try blurring my texture first...

    Note, however, that I do NOT subscribe to the ridiculous notion that MIP mapping and pixel blending should be disabled. Some newer "schools" for 3d seem to teach this as some matter of fact. IMHO, they have never done any real production and don't realize that MIP mapping is there for a reason (to prevent texture shimmer, etc.). I have gotten objects from freelancers where they turn off MIP mapping automatically (they told me to in school). These cause all sorts of problems that they then waste time and processor power fighting with AA. Most of the time texture aliasing issues are a result of disabling this feature...

  6. #51
    Quote Originally Posted by toby View Post
    This is what I used to think, but while they are post filters, they use 3d data, which you can't do in post. These two images are 1 pass aa with no adaptive, classic vs. gaussian. Guassian only took one-tenth of a second longer, and it would take several passes for classic to match it. This only works with the perspective cam, with classic cam it is just a simple blur.
    Note I would never use Classic without Oversample. (0.5-1.5 depending).

    Fine textures need more AA passes than a simple cube. The oversampled version looks comparable to the gaussian.

  7. #52
    VFX CG Supervisor Mike_RB's Avatar
    Join Date
    Feb 2003
    Location
    Vancouver
    Posts
    2,184
    mitchell is my favorite AA filter, in all apps.
    my personal company: http://www.elementvfx.com
    the people I've worked for: http://www.theembassyvfx.com

  8. #53
    Quote Originally Posted by joelaff View Post
    Note I would never use Classic without Oversample. (0.5-1.5 depending).

    Fine textures need more AA passes than a simple cube. The oversampled version looks comparable to the gaussian.
    But you see that the reconstruction filters are not just a post blur right?

    Oversample (1.5 here) is more similar to a post blur than the reconstruction filters, and took a full second longer. This image looks exactly like one with jaggies that was then blurred in photoshop. Gaussian also softens textures, and it's what they use on big-budget movies so that's good enough for me
    Attached Thumbnails Attached Thumbnails Click image for larger version. 

Name:	Ovsample.png 
Views:	161 
Size:	2.9 KB 
ID:	73721  
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  9. #54
    here's Lanczos with it's horrific white halo
    Attached Thumbnails Attached Thumbnails Click image for larger version. 

Name:	lanczos.png 
Views:	182 
Size:	5.8 KB 
ID:	73722  
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  10. #55
    Quote Originally Posted by pooby View Post
    Neither work, but Monte Carlo will work if you dont interpolate it. However its much slower than Fprime doing the same thing.
    You're right, it's phorked. Took a half hour to render a deforming box at 480 res.

    Anybody try using motion blur passes to reduce flicker? I'm having some good results with just 4 passes
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  11. #56
    Scene Destroyer DrStrik9's Avatar
    Join Date
    May 2007
    Location
    Portland, Oregon
    Posts
    1,684

    Thanks

    Hey toby, thanks for the poop on Gaussian.

    That recon filter could help defeat some noise, yes? It seems to complete the AA job nicely as well.

  12. #57
    Quote Originally Posted by DrStrik9 View Post
    Hey toby, thanks for the poop on Gaussian.

    That recon filter could help defeat some noise, yes? It seems to complete the AA job nicely as well.
    Actually I'm not sure it'll fight noise anymore than other types of aa. It uses geometry data to work it's magic, so it only works on edges. Noise and textures still need more aa.
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

  13. #58
    Quote Originally Posted by toby View Post
    But you see that the reconstruction filters are not just a post blur right?

    Oversample (1.5 here) is more similar to a post blur than the reconstruction filters, and took a full second longer. This image looks exactly like one with jaggies that was then blurred in photoshop. Gaussian also softens textures, and it's what they use on big-budget movies so that's good enough for me

    I would appear they are not purely post blur, yes. I would like to get a better clarification from Mark Granger. In another beta thread he said they were "basically" post filters. I would be interested in knowing precisely how they tie in technically.

    Oversample is not a post blur, though. According to Granger it blends the existing sample with surrounding pixels based on the current sample's position within the pixel (samples are subpixel).

    I assure you that not all movies use Gaussian... It tend to use Gaussian in mental Ray, but MR works differently, and you can control the radius of the gaussian.

    No doubt that Lanzcos and Mitchel, which both provide sharpening (i.e. white and black halos) are bad news, esp for animation. Sharpen after the render if you need to...

    Obviously different subject matter and lighting techniques require different rendering techniques.

  14. #59
    Scene Destroyer DrStrik9's Avatar
    Join Date
    May 2007
    Location
    Portland, Oregon
    Posts
    1,684

    Visible Impressions

    I never expected such participation in this thread. Thank you. I have learned HUGE amounts of stuff (not that I get it completely).

    I think I feel just a little bit cynical at this point, about GI in general, at least for animation. I still dig Except's perspective though on interpolation, and he gets great imagery, animated. So it's possible. :-) What I need to do is begin to develop some content, and learn about all the new lights (wow!), and the still-elusive node editor. Then I'll add GI in appropriate situations, when needed, and only when it works. This has been a great priority lesson for me. :-)

    You guys are awesome.

    Michael

  15. #60
    Quote Originally Posted by toby View Post
    You're right, it's phorked. Took a half hour to render a deforming box at 480 res.

    Anybody try using motion blur passes to reduce flicker? I'm having some good results with just 4 passes
    Done more tests, looks like motion blur only minimizes flicker, can't seem to get rid of it, unless maybe if it were set high enough to take much longer than other methods.

    Let's set the wayback machine for 2006, when Kray rendered this anim, with something like 8 bounces, no cg lights, no cache, in about two minutes a frame, and with deforming geo.
    http://rapidshare.com/files/238721850/DrunkenZombie.avi

    LW hasn't even caught up with that yet.
    Confirmed -
    No Weapons of Mass Destruction
    or links to Al Queda or 9/11. (Sep. 2003)

Page 4 of 29 FirstFirst ... 2345614 ... LastLast

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •