View Full Version : questions about deblock() and final encoding ...
DeathAngelBR
16th October 2007, 05:04
Ok. I got this 1280x720 HD anime. There are lots of 4x4 and 8x8 blocks and they are visible on LCD monitors, but not on my CRT unless I crank up gamma and/or contrast. Needless to say there's always someone whining about those blocks.
I'm not sure if grabbing a frame and saving it in PNG format for comparison purposes is valid. Here's the filter chain I tested:
avisource("bla bla bla")
# copy/paste/save as PNG, 282KB
deblock()
# 284KB
fft3dfilter()
# 285KB with deblock() and 284KB without
LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, smode=4, strength=125, overshoot=1, undershoot=1)
# 296KB with deblock() and 295KB without
Undot().FluxSmoothST(7,7)
# back to 285KB regardless of using deblock() or not
Questions:
Does fft3dfilter benefit from deblock, like making it easier to denoise? Would deblock() prevent these blocks from appearing on final encoding (h264) or does fft3dfilter get rid of them all?
:thanks:
R3Z
16th October 2007, 07:23
Denoisers definately benefit from having less blocks to deal with. I would suggest you use Deblock_QED (the MT2 version) as well as BlockBuster(method=dither) to help prevent the blocks (ups the bitrate a little as well). The issue is that denoising removes fine detail and then x264 removes even more and this causes the blocking issues on backgrounds or textureless surfaces.
You are not only deblocking and denoising but also sharpening. Sharpening will always re-introduce some blocking as the blocks will never be completely removed. Thats why using blockbuster (especially the dither mode) will overlay a bit of detail and mask the blocks as well.
For a test without needing to save PNG's, you can always;
Avisource("testclip.avi") #or whatever source loader
Source=Last
Deblock_QED_MT2()
(your denoiser / sharpener combo)
BlockBuster(Method="Dither")
Interleave(Last.Subtitle("Filtered"),Source.Subtitle("Original"))
Then you can flip through frame by frame comparing the result.
foxyshadis
16th October 2007, 10:21
Blockbuster is a weak salve over an issue, but it isn't a proper solution for anything. Use a matrix that won't block up to hell, like M4G Smooth or EQM AVC-HR, for that, and use a little bit of AQ if that isn't enough. If you're the one introducing the blocks, use more bitrate! It's just that simple, no need to make it look horrible by using far too little. But I'm assuming you're processing it, not outputting it.
To remove h.264 blocks, use Deblock, yes. Deblock_QED will not work for 4x4 blocking, and fft3d will partially but not fully remove them. PNG size has no meaning here, though it's decent way to quickly compare different scripts visually.
Use SeeSaw with fft3d instead of LSF, raise fft3d's strength and lower fluxsmooth's if you aren't getting enough denoising. Scrap undot all together, it's a waste of time as you use it.
R3Z
16th October 2007, 11:19
Blockbuster is a weak salve over an issue, but it isn't a proper solution for anything.
When FGM is widely intergrated for encoding and decoding i would agree with you. Until that time, blockbuster serves a purpose and it works.
DeathAngelBR
16th October 2007, 15:50
I would suggest ...
I previously used blockbuster and didn't like the results. It didn't hide or wipe out the initial blocks like deblock().
The issue is that ...
I'm aware of that.
If you're the one introducing the blocks, use more bitrate!
Hey, it's anime! I'm using plenty.
It's just that simple, no need to make it look horrible by using far too little.
It doesn't look *horrible*. The blocks appear in dark areas, which is "normal" I guess...
Use SeeSaw with fft3d instead of LSF
Tested both exhaustively and LSF just looks better in anime than SeeSaw, using relatively equal settings.
raise fft3d's strength and lower fluxsmooth's ... Scrap undot ...
FFT3D and FS' defaults are just ok. Undot is killing a few hot pixels here and there, so ...
:thanks:
Taicho
18th October 2007, 17:38
LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, smode=4, strength=125, overshoot=1, undershoot=1)
I'm curious as to why you are specifying ss_x=1.5, ss_y=1.5
when those are the default values for this filter. IE: If you don't specify ss_x=1.5, ss_y=1.5, they will take on those values anyway.
See This thread here (http://http://forum.doom9.org/showthread.php?t=84196&highlight=LimitedSharpen+settings)
foxyshadis
18th October 2007, 20:38
Actually, for smode 4, 1.25 is the default and 1.5 is probably way overkill. (Even 1.15 looks good enough usually.)
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.