GK 0.28 --> Crop by multiples of 2

[PCi]

Hi boys! Why in GKnot 0.28 i'cant crop pixel by pixel and i must do it by multiples of 2?????

[TheWEF]

because the gknot pack now comes with avisynth 2.5, which works in original dvd color space YV12 - faster, no color conversion. this also results in some limitations in cropping/resizing.

wef.

[N_F]

And on top of what TheWEF said, it's never been a good idea cropping with odd numbers. I think someone explained it with something like it will break up blocks which may give strange and bad results.

[PCi]

I know it's not a good idea deinterlace/resize a cropped source with some odd dimensions, but if you crop both sides with an odd number of pixels, you obtain a frame with even dimensions, for example weight: 720-3-5=712. And I do another time my question, why we cant crop even odd pixels if the result is even???? By the way, i think is better leave the decision to the ripper, or is really a very hard technical problem ????

Thx for your answers men ;)

[manono]

Hi-

i think is better leave the decision to the ripper

Why, if the ripper doesn't know what he's doing? GKnot's way will help you to make a better rip in spite of you trying to screw it up.

http://forum.doom9.org/showthread.php?s=&threadid=42708

[Blueseb]

all in all you can always modify crop manually in the avs script, for people that _do_ really know what they're doing this isn't a real pb

[chilledinsanity]

I agree, this sucks. Sure, you can edit the .avs file to specify the numbers, but then you have to swap back and forth between that and your preview and even then you may have to search for a bright scene so you can see how the edges are. The preview gives you a much better idea of what's going on. I can understand leaving this as default for new people, but the option to turn it off should be available, perhaps give a warning, the way it does when you use inverse telecline de-interlacing (something I do frequently). I think I'm going to revert back to the previous version...

[Acaila]

What I don't understand is why people would put up with a decrease in quality due to chroma shifts or chroma interpolations just for the sake of 1 lousy extra column of pixels? Perhaps some of you can explain that to me?

Because like Manono said, the reason GKnot forces those restrictions is to protect you from yourself, in YV12 you really shouldn't crop with odd values because you'll mess up chroma if you do. And cropping restrictions should not be optional, the way it is now you have to bypass the normal route of operations to do something that's not correct.

[manono]

Hi-

Even if you guys don't accept as gospel what I wrote and hakko504 explained in that link I gave (and I wouldn't blame you if you didn't), or even what Acaila just said (although you should), then maybe this will sound more authoritative:

http://www.avisynth.org/index.php?page=DataStorageInAviSynth

By going back to GKnot 0.27 and AviSynth 2.08, you'll be in YUY2 and the same horizontal crop rules apply. It'll just be easier to break those rules with GKnot 0.27. But you don't want to revert to AviSynth 2.08 anyway, because you'll take a major hit in encoding speed.

[folio]

Hi came across this thread cos it involves a problem i seem to be having with Gordian Knot 0.28 and the Divx Pro 5.0.5 codec. when looking at the details of the codec in Virtual DubMod it advises that the crop should in multiples of 4 =(Width) and 2=(height), whereas the WEF states that with YV12 and Aviisynth the mulitples should be 2 for width and 4 for height?!? im slighty confused by this as i never had any probs with Gordian 0.27. the films resolution came just as i wished it to when cropping(width 32) height 16).
Ive done 3 encodes so far with 0.28 and the resolution is not being calculated that well. The proportions dont seem to resize properly?! In the resolution tab of gordian ive so far put in the multiples that were stated by the WEF...but is there something im missing here...i'd appreciate very much if someone can clarify this for me? the sharpness with my 3 encodes are sharp but i think it could sharper if the resolution was being cropped correctly as in the days when i used Gordian 0.27 and Divx Pro 5.0.2...i really get on well with Gordian reckon its the best out there but i dont really want ot have to resort to 0.27 cos its slower...thanks for any help!

[manono]

Hi-

You're confusing crop with resize. The DivX5 limitations referred to in VDubMod are for the resolution. But you should still follow GKnot's W-Mod=32 and H-Mod=16 for resizing, if possible. The cropping rules are for when you cut off the black bars before doing the resize. This has nothing to do with the sharpness of your encodes, or your encodes having the wrong Aspect Ratio.

[folio]

Many thanks for the swift reply and info,

i had a doubt of some sort regarding the crop and resize...so i should continue using the 32/16 method as usual...do you think that it is advisable to check the "Enable resize" box and fill the new resize measurements in the Divx general parameters box? I only ask this cos up until ive not really used that feature with Gordian Knot...
I only ask this as im rather discerned as I seem to be getting some anomalies in all the latest encodes with Divx Pro 5.0.5, of which i never used to get so much with Pro 5.0.2?!
I use GMC and B frames and try to get the compression to 65% give or take a few values...but somehow this new codec is not as sharp...as its predecessors...is there any hints regarding Divx Pro 5.0.5 codec..up until i have religiously followed the guides at Doom and got great encodes from the Pro 5.02...seeing though that with my laptop TFT screen you cant at times appreciate how really good the encodes can be! However with the TFT at times notice the small anomalies with desktop screens dont register at all...thanks again manono for any enlightenment you might shed upon this matter!
regards
folio

[manono]

Hi-

Don't use the resizer in DivX5. Continue to do it through GKnot and AviSnth. As for the lack of sharpness, have you tried using LanczosResize yet? That might help, but at the cost of some compressibility. As for bad AR, about all I can suggest is to make sure that you have it set for 4:3 or 16:9 correctly, and try and get your AR Error down to below, say, 0.5 percent.

As for the anomalies you mentioned, you might try an encode without GMC and see if it looks any better to you. I'm not really sure what kind of artifacts you mean. Maybe a screen shot will help to explain it. By all reports, this new DivX 5.05 is pretty good, so I'm pretty sure you can work the problems out.

[jggimi]

As to Aspect Ratio: The Display Aspect Ratio (DAR) of a DVD will either be 4:3 or 16:9. But that doesn't necessarily have anything to do with the aspect ratio of the content, which, if from film, can have many different aspect ratios. The letterboxing included in the video stream will take care of that. I've seen a few "widescreen" DVDs that were transferred with 4:3 DARs.

In addition, there can sometimes be a discrepancy between the DAR in the MPEG-2 flags (reported by DVD2AVI) and the DAR in the IFO (used by players). If you can't find the correct DAR by looking at round things the resized preview window, you can always autocrop, then pick the DAR that produces an aspect ratio closest to the film transfer aspect ratio as reported on the box or at www.imdb.com -- but usually, just looking at the content is sufficient.

[folio]

Manono
im not too sure if the pic got to you. its the first time ive sent a screenshot to this forum..? hope that youve received it?!
thanks again

[folio]

thanks for the extra info jggmimi...

manono the anomalies are usually around the people almost as if it were their auras! the worst compression is noted mostly there...see what you think? thats if the screenshot has got to you?!
thanks

[manono]

Hi-

All I see is heavy Edge Enhancement. If that's true, then there's not much you can do about it. It's already on the DVD like that, and was not caused by DivX 5 (unless you oversharpened the picture with your resizer or other filter). Unfilter can help a bit, but I've found that it smooths the picture too much, before actually doing anything about the EE.

Go to this site and see if he is talking about the same thing that you are:

http://www.videophile.info/Guide_EE/Page_01.htm

[folio]

thanks manono,

will look into it...and see what the site has to say about it...cheers again
folio:D

[chilledinsanity]

So wait, you get the same problem by cropping odd numbers even if the dimensions come out even (before resizing)?

[manono]

Hi-

Yep, that's the impression I get, but I'm no expert on this. Maybe hakko504 or Acaila will chime in. But the previous link says:
and again, a 2x2 should never be split as this may create strange artifacts.
From what I gather, splitting them twice (once with the left crop, and again on the right side, for example) doesn't fix it, but can create twice the artifacts.

[Acaila]

From what I gather, splitting them twice (once with the left crop, and again on the right side, for example) doesn't fix it, but can create twice the artifacts.Manono is correct, cropping with odd values on one side is bad, doing it on two sides is even worse. Even though odd+odd=even, the chroma values won't be "repositioned" but instead you'll run the risk of cutting them off twice.

That is if chroma values are still placed in only 1 pixel out of a 2x2 pixel block, which I'm no longer sure of after reading this (http://forum.doom9.org/showthread.php?s=&threadid=52151) thread...