PDA

View Full Version : Bifrost 0.9 Released! (Temporal Derainbower)(AVS version of SSIQ too)


gizmotech
14th April 2004, 15:43
NOTE: I was did not create these filters. I'm just the pr-whore

Bifrost is a first attempt at solving the problems of derainbowing without performing spacial nuking (ala ssiq and mf). By analyzing a series of frames within a temporal threshold, it determines the max number of frames available w/o change and then performs temporal blending to remove artifacts, most namely rainbows. The filter uses a series of thresholds to determine movement on the frames, and when it identifies a progression of frames it cannot process through temporal blending (ie pans, fast motion, etc) then it launches it's alternate processor.

Which leads to the next item:
SSIQ 0.6 for AVS in yuy2 and yv12 Released!

SSIQ, previously only for vdub, was known for it's amazing powers of spacial nuking and being amazingly slow. Now you can use this amazing filter directly in avs w/o any form of color space conversion!

For comparison here are 2 screen shots. The top picture in each is the bifrost'ed source. The pic on the bottom is the original vob.
Armitage:
http://yatta.mellbin.org/misc/armitage1.png
Soul Hunter?
http://yatta.mellbin.org/misc/sh1.png

Advantages to bifrost. On slow motion/near static scenes rainbows can be greatly diminished w/o performing massive chroma channel blurring. Using the alt perameter a more powerful spacial nuker can be used on high motion scenes where chroma errors are less detectable.

Disadvantages. In some rare cases several variations on rainbowing can be too intentense to destroy w/ temporal blending alone. However when used in conjunction w/ a spacial nuker it can reduces the work required by a filter such as SSIQ and reduces chroma blurring/errors.

These 2 filters can be accessed @ the yatta site for more information.

GizmoTech PR-Whore

Mentar
14th April 2004, 15:53
These filters are brought to you by the mad lunatic who is also responsible for yatta. He goes under the name of Myrsloik, if he's not out to scare noobs from the channel ;)

You want to try this one, folks. You really do.

Mug Funky
14th April 2004, 16:55
very cool. i haven't got any rainbowed captures at the mome (i don't do much TV capture, and regularly clean my HDD of not-needed stuff), but i'll remember that this exists and recommend it to my friends :)

one nitpick - it's "spatial", not "spacial" even though the latter makes more sense when relating to "space" not "spate"...

spelling errors to me are like a non-aligned piece of paper on an empty desk. i have to straighten it :)

zettai
14th April 2004, 16:56
Ah well, better luck next time. (http://www.amvs.34sp.com/betterluck.png)

I've yet to find a derainbower that could beat this clip :)

Mentar
14th April 2004, 17:13
Zettai: Could you please make a mini vob piece available for this?

zettai
14th April 2004, 17:15
sure, I've got a whole bunch of great test clips form this dvd. I'll get them online for you.

gizmotech
14th April 2004, 17:48
I somehow doubt there is much that can be done to that clip, given the area of the rainbow effect. Seems to me there isn't much left of the chroma channel there to begin with, so restoring that w/o massive spatial(grr...) smoothing to just level everything there is probably the only way you'll get anything close to useable.

Gizmo.

zettai
14th April 2004, 19:59
As it happens the rainbow is removed on some of the frames, just not all - so there is a chance I think.

mf
14th April 2004, 21:59
Originally posted by gizmotech
spacial nuking (ala ssiq and mf).
Why, thank you for the compliment. mfRainbow has ZERO spatial filtering at defaults. The current version does do temporal nuking though, but that's a known bug.

tritical
15th April 2004, 07:28
Nice filter. I actually have a few questions that I couldn't seem to find answers to reading this thread or going to the YATTA site. First, are the pics you posted from a static scene or a motion scene (yeah its anime I know), but I mean are those scenes static for more then three frames? I assumed so just looking at them, but wasn't sure. Second, this filter will only work on NTSC material correct, since it is performing averaging between adjacent frames, or am I wrong on that? Third, it only attacks rainbows right? ..not luma dot crawl. Again I assume the answer is probably yes, but just want to make sure.
Finally I didn't see any source code anywhere to take a look at, but if you wouldn't mind could you talk a little about its basic method of operation? i.e. exactly how it finds rainbowing (for example does it simply find pixels alternating back and forth at over the variation parameter? if so... how long must the alternating be going on to be considered rainbowing?)? Does it check correspondence between filtered values at all? How is it different (on its temporal filtering and rainbow detection) from guavacomb? Can it kick in on the first filterable frame or does it take a few frames to start? I have a few others, but thats probably more then enough :) .. Again, congrats on the filter.

DarkNite
15th April 2004, 09:41
Third, it only attacks rainbows right? ..not luma dot crawl.

Looking at the Soul Hunter pics, and the test I just did myself I would say no.

@Myrsloik

Nicely done. I (and probably many others) appreciate that avisynth port of SSIQ as well. What inspired that? Are you "backing up" Pioneer's R1 Trigun? *chuckles* :D

Keep up the good work.

Mentar
15th April 2004, 10:38
SSIQ was needed to create a hybrid filter. bifrost for the "static" scenes (very much preferable to SSIQ because it doesn't cause the unpleasant "color bleeding" which is very visible), with SSIQ as backup for motion scenes where bifrost fails, but SSIQ can help out.

I think it's quite a clever solution ^_^

zettai
15th April 2004, 10:41
http://www.amvs.34sp.com/rainbows.rar

21mb rar file containing a few vob sections with some decent rainbowing. I don't have bandwidth coming out of my ass so please only download these samples if you want to use them for testing filters.

Good luck :)

mf
15th April 2004, 13:03
The rainbows in rainbow-1 appear for too short to be caught with mfRainbow. The temporalsoften that I use doesn't have any effect even unmasked.

Edit: Found the culprit! The scenechange detection in temporalsoften was too sensitive. I've now made it tweakable in mfRainbow-v0.31. But I'll detail that in my own thread instead of hogging this one. Anyway, I can now handle your rainbow-1.vob, zettai :).