Interesting Avisynth resizing thread at doom9.org
 

http://forum.doom9.org/showthread.php?s=&threadid=70734

Im at my office now, so I will check that this weekend. :)

Do use your skills to check it out! A little over my head, but it does look interesting.

In that thread, he claims: "- moving areas of each frame are softly resized and blured"

How can you apply filtering to a moving object, if there's no function in AviSynth that will give you a "sector" ( or what we already tried a while ago, by splitting a frame in quadrants) :!:
I don't see any function on that script that can apply blur to a moving object, without applying to the rest of the frame.
Or maybe I'm still not fully awake :lol:

-kwag

"I think" its something with "maskedmerge" or at least its logic.

Do you remember my PN some time ago dealing about motionmask testings?
You also answered there something about separation of the image in quadrants for motion detection.

I once did several days some testings using motionmask and MrTibs MADmask (only avs 2.0) ... it could be a way, but the thresholds of motionmask for example are very difficult to set so the Edges i.e. on static objects won't get blurry too!

hi Inc,

that sounds like a long ago snip froma thread, regarding the MA script
imrovements on motion estimates on a block-by-block bases. I think there
was talk about this long ago, in seeing if someone would create such a
filter (or .avs script) for this. Way too much reading in my days here.. I don't recall whatever became of this. Oh yes..
it was proved to be too slooooooow hehe..

I think that's what you're refering too :lol:
-vhelp

You're right vhelp :)
It's buried deep deep in this forum, when I tried slicing each frame into quadrants, and doing blur only on the parts that had motion.
It worked, but it was SLOWWWWWWW :lol:
The script I did, will be perfectly useable in the year 2010, when we are running 20Ghz CPUs :rotf:

-kwag

:mrgreen:

I knew it !! Time for more coffee and reading :roll:
Yes, the good old days of testing testing testing.. debug, debug, debug..
And, there might be some new things coming up soon :lol:

-vhelp

Yep! Its damn slow! :)

But I think the logic, to play a little with "masked edges" and performing a bulr on them would be quite interesting.
I don't think we gonna fall in love with exactly that script, but very interesting "views" in it

By changing the debug to 1 or 2 you can see which parts are motion or edge masked .... and for ma that could be a motion mask where it could perform a blur

@ kwag
Do you still have the script? Would be interesting to try it out. Even though it's adapted for the year 2010+... :wink: :D

@ incredible
How do you get the HybridFuPP.avsi to work? Could you please post your script? I'm getting an error in VirtualDub saying 'There's no function named "MotionMask"'. What values do I have to use for resizing? Now I simply put "HybridFuPP(480, 480)" but I'm guessing that's wrong :?:

Aha!
I think we need to download some sort of filter(maybe MotionMask) in order to use HybridFuPP.
But that you will find in doom9 thread audioslave.
Check it out again buddy.
C ya

Please read the thread there carefully cause I exactly got that error in vdub too, ... You need to download a special version of masktools.dll.
The download is mentioned there.

:wink:

And stay updated! As he every day releases updates of his function.

It's buried deep deep in this forum. Do a search for "quadrant", and you'll find it somewhere. I can't even remember where it is :lol:

-kwag

@ kwag
Still searching for your script... a LOT of posts to read through. :D Don't know if I have the patience to read 'em all. Thanks anyway. :wink:

@ incredible
I found the correct version of MaskTools and now it works like a charm! 8) :D

Thank you guys!

Watch his pics there and I can duplicate them!
Very sharp static parts and on the other hand a bit blurred moving parts. :D

BUt ..... its sloooow! :(

Found it :!:
http://www.kvcd.net/forum/viewtopic....=asc&start=384

Original idea was ARnet_tenRA's ;)

-kwag

New version of this function/filter: 0.6b.
Looks by faaaar better than versions 0.4b and 0.5b. Give it a spin. :wink:
I encoded "Underworld" with this and got CQ70,55 for this 2 hours movie - with 128kbps audio :!:

Did you recognise any speed-improvements? Cause thats the reason why I dropped it "till now" .... sloooow, even on my machine :(

Well, I don't really know about that... I don't think it's that slow at all. The whole movie took me about 4:45 to ancode on my AMD 2000+. Not that bad I think. :wink:

Does anyone have a link to masktools 1.4.9 because i can't seem to find one

Edit-I finally found it :D

Well I installed Fupps new Version of that function and I say:

:jawdrop:

Its faster then the old one (no probs at 352x288 and about a little less then realtime on a 480x576 ... on my machine)

To the left the treated frame using HybridFupp and to the right the original.
Its a 480x576 capture.
What I discovered is that with setting "low" much noise went away (not seen in the pic but its true) and --- I its REALLY funny --- only the motion objects did get an unsharpen kick.

Look at the Boat at the faces of the couple! And on the other hand the little details on the top of the roof of the house right... they've been kept!!
(jpeg 50% quality safed image)

http://www.digitalfaq.com/archives/i.../2004/03/1.jpg

Look at the lost details on the trees and on the mountains :roll:

-kwag

Ok Karl, but I see that in that motion scene not everything is blurred proportionally, means the whole frame !
And what makes me interesting is that "there" is a way to separate the moving parts from the static parts.

BTW: I used the preset "low" with heavier filtering on that capture cause as I said there are parts with more noise. I think at "high" setting there won't be that detail loss.

I already asked you if there's a way to implementate some "masktools" logic in MA.

What I really wonder, is if it's really necessary to blur small parts of a movie, instead of simply blurring the complete picture, on a high treshold ( very high action scene, that would mean the complete frame is in motion )
My point, if you take for example the clip you encoded (the boat), is that I can bet that the file size is just about the same on both clips, because that very small amount of blur in the very small movement parts, is hardly any to give compression over the one without the script.
So I really wonder if this filter is really usefull at all :!:
I would simply change the MA script tresholds (math) to blur at higher value, which actually means there's a lot of frame movement.
Not to mention that the speed is always going to be faster with the MA script, compared to the FuPPs script, because FuPPS calls many functions and sets many variables on every frame pass, compared to the MA script.

-kwag

After reading kwag's post I got curious and did a test on a clip of "Pitch Black" with both the MA script and the HybridFuPP (preset="High", Chroma=False).
* I encoded both clips at CQ69,9 (don't ask why...).
* Max. bitrate 2000, min. bitrate 100 (Yes, 2000 for max. bitrate value
work very well with HybridFuPP too! :wink: No macroblocks at all.).
* Clip length 2 minutes and 14 seconds.
* I used Lanczos resizing for the MA-script.
And here's what I found out:

Encoding time was about the same for both scripts, but...
MA-script > 13 499 kB
HybridFuPP > 13 192 kB :!:

Please, don't shoot the messenger. I'm just reporting my discoveries... :wink:

And that proves that the MA is bluring less, so something is being lost with FuPP ;)

-kwag

Hi audioslave,

I forgot to ask you, what were the encoding times, if you did benckmark them :?:

-kwag

Hmm,
Good question.
Say, audioslave could you give us some timings?
BTW what CPU does your PC have and what speed does it run?
Thanks buddy.

Actually, kwag, the HybridFuPP clip is clearer and sharper than the clip encoded with the MA script. If only I had some way of posting the comparison images... I don't have any webspace to upload them to.
I will reboot my CPU and do a benchmarking of the encoding times. I guess you can read that post in about 15 minutes or so.
I'll be back 8) ... Se ya! :wink:

I'm back with some info for you guys.
I encoded a new sample of the same movie (Pitch Black).
The length of the sample clip is 02:14.
----------------------
Encoding time:
HybridFuPP > 05:18
MA Script > 04:30

Clip size:
HybridFuPP > 14 419 kB
MA Script > 14 709 kB
----------------------
And once again - the clip encoded with HybridFupp is sharper...


My system:
AMD Athlon XP 2000+
256 MB RAM

That's what I thought. How :?:
MA is larger (slightly), so something is being cut off in FuPP :!:
Need some screenshots :!:

-kwag

I'd be glad to show you some screenshots but I don't have any webspace to upload them to... :oops:

www.netfirms.com ;)

-kwag

Thanks kwag! :D This is great! I'm signing up right now so I'll post the screenshots as soon as everything is alright.

So finally - here they are. The screenshots! :D

KVCD - Screenshots
http://hea.netfirms.com/screenshots.htm

As you can see the MAScript blurs the whole frame and the HybridFuPP only blur the fast moving parts of the frame. The pebbles and gravel on the bottom part of the frame is where you easiest can see the difference between these two encoding methods.

Once again - thank you for the link kwag! :D

;)

I can see it, but what I can't see is how you get a smaller file size with the FuPP script than with the MA script :!:
The FuPP JPEG file is larger than the MA JPEG file, for obvious reasons (sharpness), and that's what doesn't make sense at all when comparing the size of the MPEG files.

-kwag

I can't see the pics :cry:

Yeah, you're absolutely right about that kwag. It makes no sense but it's still true. Wierd. :roll: Have you tried the HybridFuPP function yourself yet? Looks quite amazing to me.

BTW Do the pictures in my post show this time? I can see them here at home. The last time I tried to use pictures in my post I was the only one who could see them. :) So, it would be nice to know...

EDIT: Saw your post bigggt. I'll change them to links instead. Just give me a minute...

Actually, no, I haven't :!: Yes it does :D

Edit: Indeed, if this works on every case, just like the pics you have posted, we can all say "BYE BYE" to the MA script :lol:
Then FuPP has superceded it :cool:

-kwag

@bigggt
Should be fixed now I hope. Try the link in my previous post.

@kwag
Yes. FuPP's script really makes the whole movie look sharper and "cleaner". No offence maestro. :wink:

Thanx audislave yes it works i'm looking at it right now :D