Anti-aliasing(correct term?) in rendering possible

[Path other]

Is there some kind of anti-alias setting or something to reduce the pixelated look of my finished renders in Quicktime?

 

I didn't seem to have any issue with my previous 3D program's renders which meant it must have been a default in the other program.

 

I've messed with the A:M render settings, but haven't found an answer. My outputs are set at 720x486 1 pass (4 pass didn't fix the problem).

 

Any suggestions?

[Luuk Steitner]

If you disable multi pass you have the option to use anti aliasing. For multi pass I always use at least 9 passes, mostly 16. The more, the smoother the result. But if you quicktime movie does not look good and when you render to Targa or JPEG it does, you should alter your compression settings. A bad compression setting can ruin a good render.

[Fuchur]

Is there some kind of anti-alias setting or something to reduce the pixelated look of my finished renders in Quicktime?

 

I didn't seem to have any issue with my previous 3D program's renders which meant it must have been a default in the other program.

 

I've messed with the A:M render settings, but haven't found an answer. My outputs are set at 720x486 1 pass (4 pass didn't fix the problem).

 

Any suggestions?

 

Hm... Try to render to an image (like for example TGA) to be sure it is a quicktimeproblem.

4 passes arent much so... this will generally make more or less the same output as using no multipass at all. (so just using A:Ms Anti-Aliasing-Algorithm... So if you want to be sure, use 16 passes and render one TGA... it will take long, but this is what you can get from it (16 passes are in 99.9% cases enough for anything.)

 

4 passes can be very well looking for one image but not very well looking for another...

 

*Fuchur*

[Path other]

Thanks, I'll try your suggestions.

[robcat2075]

My outputs are set at 720x486 1 pass

 

If that was with multi pass ON, that would mean no anti-aliasing at all.

 

The regular Final render with multipass OFF has the same anti-aliasing as 16 pass, multipass ON. More passes will mean better anti-aliasing but longer render time.

 

 

It's possible some pixelization is being introduced by the quicktime codec being set to too much compression, but I don't know what you've chosen there.

[Meowx]

I actually had this problem a little while back. Anti aliasing got turned off somehow and I spent about an hour trying to find the "anti-alias" checkbox... it seems to have been taken out.

 

I finally turned off the "custom" render settings and selected the default "final" setting. This seemed to fix it.

 

(correct me if I'm wrong, v15)

[NancyGormezano]

I actually had this problem a little while back. Anti aliasing got turned off somehow and I spent about an hour trying to find the "anti-alias" checkbox... it seems to have been taken out.

 

I finally turned off the "custom" render settings and selected the default "final" setting. This seemed to fix it.

 

(correct me if I'm wrong, v15)

 

You will have much more control over the rendering settings if you click the advanced option (when you go to render) and inspect all the settings - instead of using the presets. It will be much less of a mystery as to what is really set. It's also always a good habit to check the settings before rendering.

[heyvern]

If you have tight thin lines in a material, or small "speckles" like noise in a surface property this will add to any "speckly" problems in renders. That is why some projects don't need as much multipass or antialiasing as others. Another issue could be what is in the model itself. If you were rendering "THX1138" you wouldn't need a lot of multipass, smooth white surfaces... lots of bald people. If you render "Jungle Book" you would need higher mulitpass with all of those leaves and vines, hair and fur.

 

My last big project had very little need for high multipass. I was able to use only 3 steps (some shots only 2) and it still looked great (9 would have won an Oscar!)

 

-vern

[Meowx]

You will have much more control over the rendering settings if you click the advanced option (when you go to render) and inspect all the settings - instead of using the presets. It will be much less of a mystery as to what is really set. It's also always a good habit to check the settings before rendering.

This is what I normally do; after reseting it to the default I did go back into advanced. I just can't find an "anti-alias" checkbox anywhere!

[Path other]

Turning off Multi-Pass gave me the finished render I was looking for without any major change in file size and no big additions in render time.

 

Thanks for all the help!

[photoman]

I usually render my animations at 5passes but if its still jaggie I will just render it with the AA renderer.

 

Photoman

 

THX1138 Is a CLASSIC!!

[NancyGormezano]

I just can't find an "anti-alias" checkbox anywhere!

 

True - There isn't any - because - If you render with Final, No multipass - it is automatically anti-aliased. If you render with Final & Multipass & MORE than 1 pass - it will be anti-aliased - the more passes, the more anti-aliased like.

[John Bigboote]

The regular Final render with multipass OFF has the same anti-aliasing as 16 pass, multipass ON.

 

 

Really? I would have thought it would be much lower, like 4multipasses or 2.... you can see in the render dialogue with MP off as it says "antialiasing pass 1, antialiasing pass 2"...

 

For quick video stuff, I've found that there is not too much discernable difference after 3 passes. Especially if you are using 'fields', which automatically doubles your render time.

[robcat2075]

Really? I would have thought it would be much lower, like 4multipasses or 2.... you can see in the render dialogue with MP off as it says "antialiasing pass 1, antialiasing pass 2"...

 

per Martin.

 

Think of a square. Divide its length and width once and you get 4 squares. Divide again and you get 16.

[NancyGormezano]

The regular Final render with multipass OFF has the same anti-aliasing as 16 pass, multipass ON.

 

 

Really? I would have thought it would be much lower, like 4multipasses or 2.... you can see in the render dialogue with MP off as it says "antialiasing pass 1, antialiasing pass 2"...

 

For quick video stuff, I've found that there is not too much discernable difference after 3 passes. Especially if you are using 'fields', which automatically doubles your render time.

 

I believe there are 2 different methods being used for anti-aliasing when using multi-pass versus non-multipass. I don't quite understand saying Multipass-OFF is equivalent to 16 passes, although I believe it has been reported as such by others as well (including Hash?).

 

However - I thought with multipass off - that an edge detection type algorhythm was being used to smooth the jaggies. That is the first pass is computed and the second pass examines the first pass results (in the A- buffer?) detecting changes in adjacent pixels and averages them. That's why there is such a difference in computation time. In multipass - I thought a "dithering" type algorhythm was being used. Each pass takes same amount of time to compute - but imagery is offset somehow and that the passes were being averaged together.

 

It would be interesting to hear how the 2 are equivalent. ( EDIT - just saw Robcat's explanation - still don't quite get it.)

 

I agree about the 3 pass - tho I can (maybe?) see a difference between 3 and 4 pass when "soften" is also selected (for 4 pass). But I sure don't see a difference between 4 and 5 pass.

 

Maybe most differences show up if one is doing AO (which introduces a lot of noise) and one wants to blur the noise ...

 

And then ya go and compress it all - making it all icky-yuck, no matter which method.

Edited May 2, 2009 by NancyGormezano

[heyvern]

Check out the "Change Your Pants" video in my signature. Most of that was 2 or 3 passes. A very few shots... maybe 5. There might have been one shot done at 9 passes... nothing was over 9. The places it becomes obvious are motion blur "stepping". If you use motion blur you really do need more passes.

 

I too am surprised that MP off is equal to 16 passes. It doesn't seem to take as long to render without multipass compared to 16 passes. I suppose that displaying each pass might add to the time... or maybe it is purely psychological.... like when you get a new computer that seems really fast... and years later it feels as slow as an Atari computer with a cassette tape drive.

 

-vern

[robcat2075]

Here's a test scene I made to compare Anti Aliasing

 

(Shaded mode)

 

 

It's 4-patch model with a decal in the upper right that has 1,2,3, and 4 pixel-width lines on it. The model is tilted 0.5 degrees to create a severe aliasing situation.

 

 

Here's a 4x close-up of Final renders (you'll need to full size it to see detail)

 

 

You can see that the Final AA and 16 pass, while not identical, have the same number of steps of gray along the top edge of the patch and the interior of the decals are about the same. The raw edge of the decal is not anti-aliased in the Final AA version. You could solve that by clipping the edge of the decal with an alpha channel.

 

 

I think Final AA is faster because it only has to work to anti-alias things that really need it... the edges. Multi-pass is a brute force approach to anti-aliasing; every single pixel gets anti-aliased whether it needs it or not.

[NancyGormezano]

Something strange is afoot in 15e - When I render with Final (no Multi) and just the test pattern object is in the chor, it appears that no anti-aliasing is being done - but really not sure.

 

If I add another model to the chor (but make it inactive) then anti-aliasing is computed. Not sure I can see much diff between the 2. So it just may be that it isn't being recorded in the status window.

 

However there is a difference when I just do 1 pass - Not sure what to make out of all this as I believe the resultant AA in 15e looks different than the version you are using

[robcat2075]

Something strange is afoot in 15e - When I render with Final (no Multi) and just the test pattern object is in the chor, it appears that no anti-aliasing is being done - but really not sure.

 

If I add another model to the chor (but make it inactive) then anti-aliasing is computed. Not sure I can see much diff between the 2. So it just may be that it isn't being recorded in the status window.

 

However there is a difference when I just do 1 pass - Not sure what to make out of all this as I believe the resultant AA in 15e looks different than the version you are using

 

Something must be wrong because you are not getting the good anti-aliasing of decals that I'm getting in V13. In V13 the anti-aliasing of the decal is as good as a 256 pass render.

 

Do you still have 13 installed? to try?

[NancyGormezano]

Well I tryed in v13s:

 

1) I'm not seeing much dif between 13s and 15e in resultant imagery using Final (no Multi). I suspect my results are worse than yours because I grabbed the compressed jpg from the forum to use as decal, rather than starting with an uncompressed tga.

 

2) I see the same funny in 13s as in 15e, with regard to not seeing the AA passes show up in status recording window unless there is an additional model in the chor. But I do believe AA is being done anyway.

 

3) In 13s (did not test in 15e - but would expect same result) - there is a big difference if I unrotate the model back to 0,0,0 from 0,0,-5 - AA looks better if model is not rotated so severely (see image)

 

4) The best way to compare 15e and 13s would be if you zipped your chor, model, tga (or whatever you are using) - I suspect if I used your project - I would get the same results as you. But it is interesting to see what a difference rotating more severely will do, as well as using an "dirtier, compressed" decal

[robcat2075]

Well, no wonder you're getting bad results!

 

I thought i had posted the actual targa but I guess not.

 

try this...

 

AntiAliasingTestAlpha02.zip

[NancyGormezano]

I feel like I'm being set up - it's a conspiracy I tell ya!

 

Just slight differences between 13s and 15e...in AA - looks like AA is sorta similar in the 2 versions - but slightly different.

 

Obviously something is wrong. I restarted 15 - same results. I tried removing alpha channel (made decal image 24 bit instead of 32bit) - things got worse. Tried putting a basic plain white image down first then the test pattern - made no difference.

 

I'm going to sleep on it. And Let my computer sleep on it for now as well.

[robcat2075]

Those decals look completely anti-aliased, no?

[NancyGormezano]

Those decals look completely anti-aliased, no?

 

To me - ver 13s looks more or better antialiased when comparing AA Final (no multi). I guess? that AA is going on for decal in 15 - but can't really tell. As my other dirtier decal - did not get smoothed as well

 

The problem with the gross streaking in above tests in ver 15e was due to the fact that the decal didn't cover the whole face (was not a problem in 13) - so today, I added a base all white decal to the face (covers whole face) - reapplied your decal - and streaking went away. I thought I tried that last night, however I didn't realize that you hadn't covered the face, so I had just added a base white image under your image in same decal container - and therefore the problem stayed.

 

Now when I compare 5 pass NO soft to 5 pass SOFT to AA (no MP) - I notice that the soft option in MP appears to "smooth" model edges more than decal appearance (for this test pattern - not sure about others) - and that "anti-aliasing" for this test pattern looks more smoothed using 5 pass multi versus the AA, especially when comparing the thinnest line. Whether anyone cares about that is another story. But if this is an animation - one might? notice edge crawling. 256 passes was the best of all (not shown here)

 

My guess is that one has to try it out for one's particular situation (and one's tolerance or pickiness) as to what works best.

[robcat2075]

I would say the anti-aliasing of the decal in the three images you posted is functionally identical. There may be a pixel or two of difference, but it is definitely getting well-anti-aliased in the regular render.

 

If the AA of the mesh was acceptable (and this level would be in most circumstances) I'd definitely choose the regular render over the 16 pass render(4x4) render because it is so much faster.

 

 

 

The problem with the gross streaking in above tests in ver 15e was due to the fact that the decal didn't cover the whole face (was not a problem in 13)

Is that something supposed to be fixed in V15f? Did it get AMReported?

[NancyGormezano]

I would say the anti-aliasing of the decal in the three images you posted is functionally identical. There may be a pixel or two of difference, but it is definitely getting well-anti-aliased in the regular render.

 

Hmmm...it's more than a pixel or 2 - I see banding across the whole thin strip - I'm putting my nose up to the pattern and I definitely see banding, striping - However, it's perfectly acceptable to me. This just says to me that the algorithym for multi-pass averaging works well for AAing, and in some cases better than the regular.

 

If the AA of the mesh was acceptable (and this level would be in most circumstances) I'd definitely choose the regular render over the 16 pass render(4x4) render because it is so much faster.

 

16 pass? - I don't go over 5 usually - and most cases 4 is good enough. I only did 256 passes to see if there really was a difference in going with higher passes. I had never believed there was. But I proved to myself - yes there is a difference. But not worth the extra render time for a sloth like me, but might be for those who are doing higher quality still imagery.

 

And yes - no MP (regular) is definitely speedy quick, good enough, in this case in comparison.

 

The problem with the gross streaking in above tests in ver 15e was due to the fact that the decal didn't cover the whole face (was not a problem in 13) Is that something supposed to be fixed in V15f? Did it get AMReported?

 

I don't know if it got reported - I know that it's a known problem. Don't know if it will be fixed. My reports seemed to turn into the kiss of death for a fix, especially when there's a very simple work-around. I've given up on reporting. I can live with this.