PDA

View Full Version : PLD antialiasing... what's the dif?



fabmedia
09-30-2005, 11:58 AM
What is the difference between the different AA methods? I've had LW 8.2/8.3 for some time now, but I really can't see much of a difference between the various options (Box, Gausian, Mitchel and Lancoz). Is there just a speed difference with reconstruction?

I couldn't find anything on these options on the threads.

toby
09-30-2005, 12:19 PM
Hiya Fab!

If you're talking about the reconstruction filters, the only difference I've seen is that they all tend to be softer than classic, unless you choose 'sharp', where it comes out sharper, and in need of more anti-aliasing. They seem to take a little longer too.

I've yet to see or hear of any advantage to using the PLD AA types, except for the extra speed of using 3 passes instead of 5 for test renders. The quality to rendertime ratio has been worse than classic AA in every test that I've done. Extreme AA is faster than 35 pass PLD and has always looked better to me.

fabmedia
09-30-2005, 05:29 PM
I did some sample tests between pld 3 and enhanced low, and noticed that they appear to be about the same. I maybe wrong, but that's what I had noticed. Same as the pld-5 and enhanced medium. I used the Gausian reconstruction filter though.

tischbein3
09-30-2005, 06:17 PM
What is the difference between the different AA methods? I've had LW 8.2/8.3 for some time now, but I really can't see much of a difference between the various options (Box, Gausian, Mitchel and Lancoz). Is there just a speed difference with reconstruction?

I couldn't find anything on these options on the threads.

reconstruction filters (renember: imho)

at high pass values they all start to get close to classic at low there is a lot of difference:

Lanczos: like classic + running any sharpening filters over them.

Mitchel :
although close to classic, antialising jittering is greater when it comes to bump and small details.

Gaussian: I compare it with lower classic plus blur

Box:
Quite similar to gaussian, although I've experienced better results (sometimes even better than classical) when it comes to high constrasted areas / vertical / horizontal lines and bump.


PLD (also very imho):
PLD is somewhat faster, resulting in more "sharper" object edges.
Very critical when it comes: to bump or small detail , close to horizontal / vertical lines and DOF

Just personal opinion, best thing to do is to do
your own experiments on this.

fabmedia
09-30-2005, 06:23 PM
-- at high pass values they all start to get close to classic at low there is a lot of difference:

You'll have to restate this as I don't understand that.


When you say "lower classic" do you mean Low/Enhanded Low or are you including Medium?

tischbein3
09-30-2005, 06:43 PM
Yep desperatly need sleep...

An example:
lets say you use pld-5 pass with lanczos and compare it with
classic medium, there is a significant difference between both.
(As said, Lanczos has a more constrasted /sharpening result)

When you do now higher the passes of Lanczos the result are getting more the look wich is comparing to the result of classic medium. Becuase more passes were used it will look better, but in general it starting to get more the typical classical "look" (image will not look sharper).

fabmedia
09-30-2005, 07:47 PM
Have you found that if you render with classic and then 2 passes lower with PLD, that look the roughly the same?

Integrity
09-30-2005, 09:47 PM
I could be totally wrong about all of this, but I think the PLD stands for Programmable Logic Device and the names are just the different functions in defining the curve in the convolution process. I think Newtek named the reconstruction filters after this.

I thought one difference between PLD and Classic was that PLD calculated the center sample fully and only calculated the geometry aspect of the other samples like graphics cards do in their different AA modes, which makes it render faster compared to the Classic 5 pass. But reading further on the internet I think it's just using the neighboring aliased pixels to calculate the anti-aliasing using a PLD's function of reconstruction. Box being a sort of cut band pass, gaussian being a nice bell (which makes it look like a blur), and lanczos being somewhat of a diffraction curve (which makes it sharper). I don't know what mitchell is though.

Lynx3d
10-01-2005, 08:20 AM
I think the PLD stands for Programmable Logic Device
*lol*
no, it stands for "pixel lattice deformation"
But Newtek doesn't really tell what's behind that, obviously a different way of choosing sample positions than the classic AA does.


I thought one difference between PLD and Classic was that PLD calculated the center sample fully and only calculated the geometry aspect of the other samples like graphics cards do in their different AA modes, which makes it render faster compared to the Classic 5 pass.
;D


I don't know what mitchell is though.
Mitchell and Netravali proposed a configurable cubic spline filter, though many renders use a fixed configuration with slight negative lobes to find the best compromise between gaussian and lanczos (wich precisely would have to be called lanczos windowed sinc...)

fabmedia
10-01-2005, 09:57 AM
Mitchell and Netravali proposed a configurable cubic spline filter, though many renders use a fixed configuration with slight negative lobes to find the best compromise between gaussian and lanczos (wich precisely would have to be called lanczos windowed sinc...)



Huh?!?!?!?

trick
10-01-2005, 10:55 AM
Arlen,

I really think you're a very nice guy, but please lose the habit of posting threads with the same topic in more places. You may get a faster reaction but it is hard to keep an eye on all replies...
Thanks !!

fabmedia
10-01-2005, 11:08 AM
Sorry man. I wasn't getting any response at all with the first post, hence I went to the Mac forum. I'm slightly impatient at times.

I appolgize.

fabmedia
10-03-2005, 01:25 PM
Okay, here's some previews for differences.... model is Christopher Dessey's.