another restore function

[Adub]

Do you have an example? Preferably an example clip, or screen shots if we have to.

[MOmonster]

Sorry for the delay, but there are some other things I have to do.


@the_toof
Here is a simple function to solve the chroma problem. I was wrong it is indeed blending. Clear chroma fields are missing, but this is fixable:

see page 19

Just bob the source and add this function. Decimation is still necessary.

Your second example (the music clip, I think) is to messy for a simple automated solution. Some parts have to be treated as interlaced, other as progressive, combined with many artefacts and different framerates. In my opinion it`s really not worth the work.


@EuropeanMan
I don`t know why, but I can't download your sample from sendspace.


@Piper
As Merlin7777 allready posted, without a sample I can only guess.

[thetoof]

Many thanks!

[Piper]

@Merlin777 & MOmonster: While readying a sample clip, I discovered that by increasing the dmode & omode parameters to 8, the issue resolved itself.

The source clip I had intended on sharing included a scene change followed by 2 blended fields. Using srestore defaults, the same 2 frames appeared a bit blurred, the 3rd being clear. With the tweak, transition is smoother.

Thanks

[MOmonster]

Maybe a bug in this version, but I have allready finished the next release.

Update
I uploaded the srestore 2.7 prerelease.

[Adub]

Changelog?

[MOmonster]

ok, short changelog:

- removed some parameters (pclip,pthresh,chroma,... see readme)
- added possibility to merge duplicates
- fixed a bug in the double blend removal
- significantly saver detection for smoother motion
- new blend detection
- code cleanup

I'm thinking of some further changes, but on the less sources, srestore didn't perform good, the new version improved allready a lot.

[elguaxo]

thanks MOmonster!

[thetoof]

I have a hard time removing the double blends in this source... could you recommend some settings?
Small mouth blends are kept and there is a mess up after the scene change. I'm using the latest srestore.

[MOmonster]

Very bad source.
I see the problem:
The motion of the hair is much bigger than the mouth motion. That's why srestore detects this as a single blend.
Using other settings won't help. I'll try to improve this in the next version.

[thetoof]

Originally Posted by MOmonster

Very bad source.

Yeah... maybe some prefiltering would help?

[MOmonster]

I had a closer look on the source. The detection will be saver in the next version, but the two problems with the sample won't be solved.
1. The scenechange is also an edit. So the first clear frame followed by the double-blending isn't there
=> a right and save detection isn't possible in this case
- next version won't touch these frames
2. The text that running through the picture is not blended and the lumadifference of the scrolling text is higher than the difference caused by the mouth-motion. That's why the detection also doesn't work in this case.
=> This is not restorable. The text will be messed up.

[MOmonster]

Srestore has reached version 2.7c

- changed many internal conditions to improve detection
- same speed as v2.6 (a little nit faster than 2.7pre)
- changed double-blend-detection
- no changes in usage


@thetoof
The detection works better, but as long as you don't crop the text, you won't be satisfied.

[thetoof]

Goodie good, I'll test it tonight by using a cropped dclip.

[ministrix]

i got troubles removing the blending in this clip. Tried different settings on srestore but didnt remove all blends. Is there anything i can do to tweak it? the dvd is from UK but its ntsc so my guess its a badly converted job. Ive used srestore several times and very pleased with the filter so tx a bunch momonster for a great job

[MOmonster]

The sources is not fully restoreable. Some blends aren't removable.
To set the right output framerate is very important for the pattern detection of srestore.
These settings give significantly better results than the default ones one your clip:

srestore(speed=-9,frate=23.976)

[thetoof]

All the links in this post were posted previously in this thread... it's just to reduce the possible confusion due to referring to many clips.

-Double blend removal works like wonders
-Chroma_rep leaves some of the blends behind (check frame #4 of the Simpson sample)
-Could chroma_rep take into account only chroma blends (i.e. if the blend is a "luma+chroma blend (normal blend)", don't consider as a blend)... this could make automatic restoration of chroma blends work on the music video sample and then srestore could be called to remove the simple blends & decimate accurately (the chroma blends srew up the detection).
-There was no answer about this issue

Thanks for everything you've done so far!
toof

[MOmonster]

Thanks for remembering, I just lost the overview.

- good to hear
- will have a look on it, function is definetly improvable
- if chroma_rep and srestore are necessary, chroma_rep has to be used before. Chroma_rep detects only this special case of chroma_blending. that have nothing todo with the luma channel, and srestore doesn't regard the chroma with default settings, so there is no reason why chroma_rep or chroma blending in general screw up the detection of srestore (maybe problems with the conditional enviroment).
the decimation of srestore won`t work on your clip. There is no possibility srestore can work nice on hybrid sources. With a special more agressive blend-deduping blenddetection, it would be possible to remove most blends in this clip, but I don`t know if this is worth the work, this is the only example I know with this strange blending.
- I downloaded the sample and will have a look on it later.

Edit: I had a look on the source and this is not restorable. Triple-Blends and more. Maybe a normconversion that was just deinterlaced. Only solution is getting another source.

[18fps]

Originally Posted by MOmonster

To set the right output framerate is very important for the pattern detection of srestore.

Thanks for your great work, MOmonster! I use your script mainly to deinterlace / deblend silent movies, where figuring the right speed is (to me at least, as I get lost after a couple seconds of counting unique frames) the most difficult part. On this particular source, however, your Interlacing2reader tells me there are some fields shifted. I should pre-process this source before calling srestore? Also, what parameters of srestore would you recommend considering that I prefer a blend than a duplicate ? (framerate should be 19 fps). Thank You!

[MOmonster]

There are't enough blends in your sample and the decimation ration is higher than 3, so srestore doesn't find a pattern. Using srestore as deblender and add the right decimation works better:

bobbing
srestore(omode=4,cache=5)
tdecimate(cycleR=41,cycle=60,mode=1)

This is the decimation ratio, I found, but it doesn't have to be the the right one (small changes are possible).