sampling problem in blurs

[spetz]

good day fellow wizards!

I have a encountered a problem several times while trying to use wide offsets in Gmasks which is located on a chromatic bkg, especially when it contains a certain amount of gradient, (differences in hue levels). it seems that the bigger the blur is, the less sampling the offset reach. and you get a stroby looking offset, which in you can clearly see the grading of it. (like bars of gradients). I am working in a 8bit environment, but I am not sure that is the reason, and if it is, so how can I make it smooth looking again, and achieve the best results? in short: please help. 🙁
spetz

[Keyser_Soze]

spetz wrote:

good day fellow wizards!

I have a encountered a problem several times while trying to use wide offsets in Gmasks which is located on a chromatic bkg, especially when it contains a certain amount of gradient, (differences in hue levels). it seems that the bigger the blur is, the less sampling the offset reach. and you get a stroby looking offset, which in you can clearly see the grading of it. (like bars of gradients). I am working in a 8bit environment, but I am not sure that is the reason, and if it is, so how can I make it smooth looking again, and achieve the best results? in short: please help. 🙁
spetz

If it’s FFI you are talking about. Try and blur the matte channel and hit the gaussian blur button. Cheers.

[spetz]

I wouldn’t have posted this Question unless I have tried this simple, and let me add; Logic solution. adding blur to it only make it worse. the blur is what causing it in the first place… it looks like the software have some kind of difficulty calculating it, resulting this artifact. I was thinking it’s got to do with the 8bit inv’ I am working in. I guess the solution will come from someone who is more in to the technical part of comp’ softwares. (not necessary FFI technician). but I appreciate the reply, thanks anyway.

[Keyser_Soze]

spetz wrote:

I wouldn’t have posted this Question unless I have tried this simple, and let me add; Logic solution. adding blur to it only make it worse. the blur is what causing it in the first place… it looks like the software have some kind of difficulty calculating it, resulting this artifact. I was thinking it’s got to do with the 8bit inv’ I am working in. I guess the solution will come from someone who is more in to the technical part of comp’ softwares. (not necessary FFI technician). but I appreciate the reply, thanks anyway.

Well, yes I guess it was a simple solution, sorry it didn’t help you. Wide offsets on gmasks always tend to look a bit ugly in FFI so it’s quite often that I add a blur to the matte channel to make things look better. Apparently that wasn’t the problem this time.
This is no solution either but it will help you find out if it’s the 8bit thing or not (and I guess it’s kinda obvious as well so you might have tried it already?). Anyway, just export your source elements and recreate the comp in a piece of software that can handle higher bit depths, like shake, fusion or nuke. And talking about gradients it’s always a good idea to use grain to reduce banding (yes, simple trick too :-)) . Merry christmas and good luck.

[loops]

Try this simple test 🙂

Zoom right in on your comp and set up a reference split so you can also see the output of the GMask node. Grab a colour pot from the Guides section (or anywhere else), set it to RGB Col mode, hit pick and have a drag around the GMask output. Are the patterns you see caused by variations of 1 in the mask colour? And are the patterns you see whilst zoomed in on the GMask the same as the ones you see in your comp? If so then you’re seeing quantisation noise from working in only 8 bits, so, rinse and repeat in 12 bit graphics. Can you still see patterns? Are they still 1-bit high? Don’t try this on an LCD monitor, they’re all 8 bits. And keep Use Ratio off, it confuses the issue.

You don’t normally see quantisation noise with “normal” footage, but it is pretty obvious if you do a huge horizontal fade over something very CG in 8 bits. The GMask node doesn’t add dither, which is fair enough, it slows things down and you can always add it yourself – Regrain with height and width set to 1, and a sorta triangular shape with height about 1. At least that’s the mathematically correct way to do it, but don’t feel you shouldn’t do whatever looks smoothest.

What annoys me about GMask is the way things fold over and go weird if you have a tight shape and you turn the offset up too much, without using Advanced Gradient.

[Anonymous]

as loops is pointing out, you should first check out & analyze the steps of the gradient, if the banding corresponds with these values, your only option (besides 12 bit) is regraining or other ways of dithering.
if not (more color sample steps than visible bands), you could try with 2 overlapping different scaled gmasks each with a different offset … (one pos, one neg)

good luck

robert

[spetz]

Cheers for the complex and interesting answers, I will get right to it.
merry Christmas you all!

[Keyser_Soze]

loops wrote:

Regrain with height and width set to 1, and a sorta triangular shape with height about 1. At least that’s the mathematically correct way to do it, but don’t feel you shouldn’t do whatever looks smoothest.

I didn’t really follow you there… what do you mean with “triangular shape with height about 1”?

[loops]

Lucky you asked that because I looked it up and it actually doesn’t work at all how I thought it did 🙂

If you pick the Custom stock type in ReGrain, you want flat RGB curves in the little curves window you get, not triangular ones at all 🙂

[spetz]

hey loops,
I didn’t want to post a new topic just for this simple Q, so I haded it to these subject if thats ok..
I wanted to know how can you print screen, like a schematic in Action/batch in a unix env. the com is obviusly conected to a network, which is also conected to pc/mac and printers if that helps.
oops…almost forgot…on Flame…hhh
spetz

[loops]

If you’ve just got normal Irix with no extra bits installed, you can use mediarecorder – run it from a shell. It has a fairly strange GUI.

Or you can use scrsave from the shell – to wait for five seconds, then save the whole screen to screen.sgi as an SGI image, in the current folder:

Code:

sleep 5; scrsave screen.sgi

The five second delay should give you time to hide the shell window away 🙂 I have to confess I’m not quite sure how you’d go about doing it in 12bit…

[Author]

Posts