IIP/Limited Sharpen combo problem

[Socio]

I know Didee said there will be no support for the special IIP/Limited combo but I figured I would go ahead and point out a serious bug in case Didee was incorporating it into the "Sharpening Suite" that we are all anticipating, well at least I am and was unaware of the bug.


Anyway I found that if you enable Supersampling in any form it kills the sharpening part of the process.

The following are three pics one is normal, one with Supersampling set to 1.5 (does not matter even 1.1 will cause the bug) and one with Supersampling set to 1.0 which is off.

Normal

http://img31.exs.cx/img31/1697/magnorm13jc.jpg

Supersampling set to 1.0 (off)

http://img31.exs.cx/img31/6604/magfull16fz.jpg

Supersampling set to 1.5

http://img55.exs.cx/img55/6622/magorig14ff.jpg

[Didée]

Ooops. I know of one bug in that inofficial iiP version, but it's another one: If you specify "dering=[positive value]", then the protection mask gets completely white, thus disabling all sharpening except the SS-LS.
Could it be you used a positive value for deringing?

Anyway, thanks for the bug report. I'll have to crosscheck what's actually going on.

(BTW, the announced "sharpening toolkit" has nothing to do with iiP, not even with its working principle. It's just LimitedSharpen, with some minor :D additional features.)

[Socio]

Thanks for the response,

No, I am using -60, maybe it is the reverse and I need a positive number instead; I will experiment with it later today.

[Socio]

I tried using a positive number but it gave me a "unfilter" error. However when I downloaded the Unfilter plugin and loaded it in my call script and again used 60 instead of -60 I did not get the error but the image was still the same. I tried -60 with UnFilter loaded and that did not make any difference either.

I even hacked IIP and cut out the LimitedSharpen part and ran LimitedSharpen full version as a separate call along with IIP and the picture still came out the same if you enabled SS in IIP.

I will mees with it again later.

[Didée]

Hmh. Right now I can't look at it - perhaps later today. But looking at the script, I can't see anything related to SS that could trigger an error: that's only about either to place, or not to place a resize() in the middle of a linear chain. (And the "dering-bug" was only in one of my devel versions, not in this one.)

Could you please check with >> debug="showall" << if any mask, especially the protection mask, turns into fully-white when that error occurs for you? Thanks. (I recall that on rare occasions, the Ylevels() function all of a sudden gives nonsense output ... but only when used within iiP...)


EDIT:

Nope. That version works correctly for me, with and without SS. Either some other parameters are set too strong for you, or it's a problem with a plugin.
Try setting "protect_bias=1" and "dering=-1", that should give (waaay too) strong enhancement. If not, something other is wrong.

[Socio]

Ok, I think you maybe right about it being a plugin because I also tried the original IIP with xsharpen at default settings and got the same results.

I removed, and re-installed all plugins needed and am running another test to see if it fixes it. I think it might be the MPEG2Dec3dg.dll that I installed as part of the DVD Rebuild installation via the DVD-rebuilder guide's instructions, not 100% sure yet.

[Didée]

Got it.

With the LS integration, the 2nd SS factor is much smaller, and therefore the "subpel" smoothing must be disabled ("subpel=0.0"). I wrote that in the comments.txt file (and the "default function call" example as well), but forgot to adjust the default setting in the script. Therefore, when subpel is not explicitely set to zero, it will be done with the wrong default value ("blur(1.58)"), thus destroying all formerly done sharpening ...

Also, for some reason the "protect_bias" value should be lowered ... 8 to 12 seems to work more reasonable than 16.

[Socio]

Adjusting the subpel to 0.0 and lowering the protect_bias from 20 to 12 worked like a charm. I can use SS and get even more detail than before.

Thanks Didee!