BlendBob 1.2: smartbobbed-field matching and de-rainbowing filter

[Chainmax]

I'll wait for the new version before jumping on the bandwagon, but after reading this thread I still have a doubt. From what I can understand, this filter has to be followed by decimation in order to achieve complete IVTCing. Is that correct?

[Leak]

Originally posted by Chainmax
I'll wait for the new version before jumping on the bandwagon, but after reading this thread I still have a doubt. From what I can understand, this filter has to be followed by decimation in order to achieve complete IVTCing. Is that correct?

Sure - my filter takes a video that's been bobbed to 59.94 FPS and blends it down to 29.97 FPS again, but this time combining every second frame with the neighbour that matches it most.

This means that in a properly telecined stream you'll get one of the shifted fields matched with it's predecessor (which will match it better, unless there's absolutely no motion going on), so you'll also get a duplicated frame every 5 frames in the output (just as with Telecide), which Decimate will happily eliminate.

What you shouldn't get is residual combing (unless your bobber did it's work poorly), as a blend of 2 frames made by deinterlacing fields can't introduce extra combs.

np: Burnt Friedman & Jaki Liebezeit - Obscured By 5 (Secret Rhythms)

[Chainmax]

Originally posted by Leak
Sure - my filter takes a video that's been bobbed to 59.94 FPS and blends it down to 29.97 FPS again
I'm sorry if this this is an ignorant question, but wouldn't that introduce ghosting in the final output?

Anyway, how feasible would it be for you to incorporate a decimating function and make this a full IVTC/derainbowing package with each function accesible separately? I know I'd love to have to use YATTA only for VFR encodes and not pattern matching correction.

[Leak]

Originally posted by Chainmax
I'm sorry if this this is an ignorant question, but wouldn't that introduce ghosting in the final output?

That totally depends on the input - my plugin always blends the 2 input frames that are most similar; if you feed it telecined or real 30 frames per second material, it should blend each frame with the frame that resulted from deinterlacing/bobbing it's other field, you shouldn't get ghosting. For telecined material, you'll get a duplicated frame every 5 frames, but no ghosting because of that.

If you feed it 60 fields per second material, like a sport broadcast or similar, you will get ghosting, as there really are no matching frames, but it'll choose one of the two possible anyway.

And by the way: is there some reason you always ask about each and every thing when just doing a small test would yield the same answers, and quicker at that?

Anyway, how feasible would it be for you to incorporate a decimating function and make this a full IVTC/derainbowing package with each function accesible separately? I know I'd love to have to use YATTA only for VFR encodes and not pattern matching correction.

The problem with making the functions accessible separately is that you can't pass anything else than video or audio data around in AviSynth, while I'd need to pass other hints between those separate functions you propose. I'll probably also make a decimation function someday, but it'll probably end up inside one big function that does everything internally...

np: Radiohead - I Am A Wicked Child (Com Lag (2+2=5))

[Piper]

I've been toying with DGBob and KernelBob in combination with BlendBob for the past few days against pure interlaced sources (NTSC). I can't seem to get around one issue - motion always appears more fluid when using DGBob than with KernelBob. I'd rather have KernelBob be my default/preferred option - esp with Leak's recent MMX optimizations (THANKS btw). ;)

Can anyone help improve this? This is my script as it appears now:

# HUFFYUV YUY2 Source
AVISource("d:\capture\capture.avi")

# Crop when downsizing to 352x480 from 720x480
Crop(8,0,-8,0)

# Bob
#DGBob(1)
KernelBob(order=1,threshold=8,sharp=false)
ConvertToYV12()
BlendBob()

#Denoise - added at the end for speed reasons alone. ;)
CNR2()
UnDot()
RemoveDirt()
LanczosResize(352,480)
Blockbuster(method="noise",variance=0.3,seed=1)

[Chainmax]

Originally posted by Leak
is there some reason you always ask about each and every thing when just doing a small test would yield the same answers, and quicker at that?
Sadly, there is. My computer is a crappy 850Mhz T-Bird with 128Mbytes of PC100 ram (originally 256Mbytes of PC133 ram, long story) that acts even slower than it should due to a host of problems that will basically only be solved by an upgrade which won't be feasible for several months due to financial reasons. On top of that, it's the only computer in my house and it's used by at least two other persons, leaving me with little to no time for me to use it. Finally, I only have a few DVDs and no capture card, therefore I don't have the necessary range of sources for a proper test of any kind. Really, even though I am quite the procrastinating type, most of the questions I ask are due to the fact that I just can't solve them myself. I apologize if I annoyed you :(.

[Leak]

Originally posted by Leak
(about my np: lines)

I guess you'll be completely disappointed, but I do that by typing it by hand... :D

I've just converted to the dark side, I mean foobar2000, so from now on my now-playing lines will be done by hitting CTRL-ALT-SHIFT-C followed by CTRL-V... :p

np: Yoko Kanno - Monochrome (Ghost In The Shell Stand Alone Complex OST 1)

[Chainmax]

I couldn't wait until the new version is released, so I tried the current one on my Spawn DVD. This filter is awesome, unlike Decomb or SmartDecimate there is no remaining combing and there's also no "color ghosting" as there is when blurring the chroma channel after Decomb or SmartDecimate. There are some jagged edges here and there and some minor ghosting, but overall the result is much more pleasing through the eye (at least when stepping through it in VDubMod). Here are the settings I used:

KernelBob(1,5,false)
BlendBob()
Decimate(cycle=5,mode=2,quality=3)

[jarthel]

completely ignorant question: What's the equivalent of the following scripts to blendbob?
--------------
Telecide(order=1,guide=1)
Decimate(mode=3,threshold=3)
----------------
Telecide(order=1,guide=1)
Decimate(mode=1,threshold=3)
-----------------
FieldDeinterlace()

[Zarxrax]

Sorry if this is a stupid question, but I only just recently discovered this wonderful filter! I was just wondering if LeakKernelDeint is still the preferred bobber to use with this? I tested the kernel mode in TDeint with blendbob, and the result was a little more compressible than with LeakKernelDeint, though it was a lot slower. Visually, I couldn't really tell much difference between the 2. There are a couple other modes in TDeint that I couldn't really tell much difference between either.
Should I just stick with LeakKernelDeint for its nice speed, or does another method offer some advantages that I might not be seeing? Source is anime, by the way.

[Leak]

Sorry if this is a stupid question, but I only just recently discovered this wonderful filter! I was just wondering if LeakKernelDeint is still the preferred bobber to use with this?
To be honest, I only ever tested it with Donald's KernelDeint, DGBob and simple Bob - but as long as it's a bobber that preserves static areas it's probably going to work with whatever you throw at it...

Then again, I haven't used this filter for quite some time thanks to tritical's TIVTC, so experimentation might yield even better results than asking me... :)

np: Jah Batta - Music Hit You (Rhythm & Sound With The Artists)

[MacAddict]

Depending on the source, I find that DGbob also provides a little better fluid motion with BlendBob. It's usually trial and error for each source. I need to throw Tdeint's bobber in the mix on some testing too :-)

[acrespo]

Leak: BlendBod derainbow feature is great. Can you separate this option in other plugin?

[psme]

AMAZING! AND THANKS!

This filter (together with LeakKernalBob) BEATS ALL other 3/2,2/2 pulldown deinterlacers (both software/hardware) on my problem clips! On my 84" projection screen, the image is simply PERFECT!

Thanks for bring back this thread otherwise I may never notice!

How can this be SO PERFECT? Now I need to re-watch all my shoot on 24/30fps film MTV DVDs which have tons of bad edits and encoded as 60i!

I still feel SCARY! :D

regards,

Li On

PS: realtime DVD playback only, ZoomPlayer, DScaler5 decoder, FFDShow->Avisynth interface.

[MacAddict]

haha...sorry man, still trying to find some good solutions for field blended 'films' :-) Glad it worked for you though!

[psme]

I know! I have a few Anime DVD movies (from Bandai) I think in the "field blended film" format. This BlendBob won't help, as expected. I tried a few other blended film filters but nothing seems to run in realtime in FFDShow->Avisynth and I don't do encoding!

regards,

Li On

[acrespo]

psme: The best way to IVTC my animes is:

Leakkernelbob(1,5,false)
BlendBob(derainbow=1)
TDecimate(mode=1)

TFM can't reach the quality of blendbob because the lines are jaggy too much.
Leak recommend TFM but the best for me is BlendBob!

[Leak]

Leak: BlendBod derainbow feature is great. Can you separate this option in other plugin?
Well, I remember that I didn't separate de-rainbowing from the rest of the plugin on purpose, but it's been some time since I last worked on it.

Still, from the description I read about Bifrost the de-rainbowing works quite the same in it, so why don't you give that a try? :)

Alternatively, you could set BlendBob to just spit out the rainbow mask and blur the chroma in the masked parts on your video (as that's what I'm doing in the rainbow-y parts).

Alas, there's one problem here: the rainbows are detected by comparing the (bobbed) chroma of the two fields that will be blended by BlendBob (that's why it only works if the rainbows were introduced on a per-field basis), so if you use something else to IVTC your video the rainbow mask might be off.

I know I probably should add an option to just weave frames if they're similar enough, but I hardly ever get time (or motivation) to code stuff lately (doing it at work for 8 hours a day makes you really want to do different things in your free time) - not to mention that I've been using tritical's TIVTC for ages now... ;)

np: Markus Guentner - Droid Ships (Audio Island)

[MacAddict]

Wish I understood TIVTC a little better. I've got way too many Asian type films with blends and weird stuff going on :-(

[Zarxrax]

Edit: Nevermind, stupid question.