DeblockPP7

[HolyWu]

https://github.com/HomeOfVapourSynthEvolution/VapourSynth-DeblockPP7

[HolyWu]

Update r2.


Add SSE2 code path.

[ChaosKing]

I get white dots on black edges in many scenes:

clip = clip.pp7.DeblockPP7(qp=5)
https://i.imgur.com/lfKCGvC.pnghttp://

Happens in r1 & r2.

And as always awesome work! :thanks:

[HolyWu]

I get white dots on black edges in many scenes:

That's weird. I use your picture for testing but can't reproduce the same issue. Anyway, try DeblockPP7-test.7z (http://www.mediafire.com/file/355vygaynln97dg/DeblockPP7-test.7z) and see what happens.

Besides, the source seems not blocky...or I'm blind.

[ChaosKing]

That's weird. I use your picture for testing but can't reproduce the same issue. Anyway, try DeblockPP7-test.7z (http://www.mediafire.com/file/355vygaynln97dg/DeblockPP7-test.7z) and see what happens.


No more white dots.

It's blocky in some other scenes, I just wanted to show the bug.

[HolyWu]

No more white dots.

Thanks for your report.

Update r3.


Fix pixel overflow.
Add SSE4.1 code path.

[ChaosKing]

Quick bench on ryzen @3.7ghz DVD 720x480
Deblock-text: 750fps
r3: 860fps

This deblock filter is fast!

[Selur]

Thanks :)

[Boulder]

Is the filter more suitable for cartoon or real life sources? I've tried to find information over the internet but I didn't see any tests.

[HolyWu]

Is the filter more suitable for cartoon or real life sources? I've tried to find information over the internet but I didn't see any tests.

I can't say whether it's more suitable or not. If you have some blocky DVDs or videos at hand then you could just try it by yourself.

[Boulder]

I tested it briefly on a MacGyver DVD that I had on my HDD. The filter is very strong and blurs the image noticably already at qp=2 so it seems that for real-life sources, I should still use SmoothD2 with the Avisynth proxy.

EDIT: Thanks anyway for a port, there are never too many of them :)

[HolyWu]

I tested it briefly on a MacGyver DVD that I had on my HDD. The filter is very strong and blurs the image noticably already at qp=2 so it seems that for real-life sources, I should still use SmoothD2 with the Avisynth proxy.

May I ask you to provide an unprocessed video of 5~10 seconds for me, also the arguments you used for SmoothD2, hence I can have proper materials for testing. Thanks.

[Boulder]

Sure - here's one clip to test on: https://drive.google.com/open?id=1rsQXAZ4e6FUxj5kPFYMPloQrPfd-Mtzz .

I would probably use SmoothD2(quant=5, zw=3, num_shift=4) on the episodes at least based on this clip. It affects the noisiest areas, pressing them down slightly but doesn't oversmooth the few details that are left. Of course, some people prefer a smooth output but in my eyes, a slightly noisy video is better to watch.

[HolyWu]

Update r4.


Change default qp value to 2.0. The type is float instead of int now.
Add mode parameter. Previous versions use mode=2 internally.

[Boulder]

Thank you, it's very usable now with the fine-grained qp :)

EDIT: A nice speedup over SmoothD2 too, 450.99 fps vs. 25.54 fps :D