x264 encoding in 10bit 4:2:2 for Blu-ray

[TheDock]

Originally Posted by Ghitulescu

I noticed we were discussing SD, not HD, but with VHS or S-VHS even 8b is too much. Capturing them in 10b or more is a waste of resources.

Thanks for your comment, but I was not asking for this kind of "help": if you read my question, it is not mentioned that my workflow is for VHS only. And actually: even capturing video with 1bit is a waste of resources for a blind man that does not require post processing. Further, you should have a look at the recommended archiving methods of analog supports. Please, also think about the fact that the required "resources" today (in this precise case storage space) will seem low in 20 years (and I have no problem with storage space currently, but thanks for taking care about it for me).

One last detail: I am working at Fraunhofer...

[pcordes]

Originally Posted by TheDock

I totally agree that the bitrate it is not "needed" in this special sample, but as I said, this was not my question.

I was just curious what got you interested in playing around with super-high bitrates for SD.

"..\ffmpeg\ffmpeg-20150206-git-9dc45d1-win64-static\bin\ffmpeg.exe" -y -analyzeduration 100M -probesize 100M -ac 2 -channel_layout stereo -i input.mov
-vf crop=w=700:h=560:x=20:y=4,pad=width=720:height=576:x=12:y=8:color=black,fieldorder=tff
-vcodec v210 -acodec copy output.mov

lossless libx264 (-preset ultrafast -qp 0) will use a lot less disk space, and is quite fast to encode and decode, thanks to ASM-optimized decoders. I'd recommend using it over ffvhuff or utvideo for lossless scratch files, if you want to compress your output.

Concerning the conversion to 50p, I am working on it

[/quote]

I recently deinterlaced a DV source (which I only had a DVD copy of :/). I got quite good results with

-vf yadif=3:1,mcdeint=3:1:4

. It ran all night on my sandybridge (on 720x480. single threaded...), but the output looked good. Dunno how much better it was than I would have got with mcdeint=2. (The last parameter is not a higher-is-better, though. Read the docs to see what it does.) The 2nd parameter (to yadif and mcdeint) is 1 for BFF, 0 for TFF.


Since you're on Windows, you probably already have AviSynth set up. (I tried on my windows machine, but I grabbed 64bit avisynth+ so I could use the 64bit x264.exe I'd already cross-compiled, and then found out that some avisynth plugins are only available as 32bit code.)

Anyway, apparently QTGMC is the best available free (but not open-source?) deinterlacer. And it's still stuck on 32bit only. Next time I have to deinterlace something, maybe I'll find a way to pipe data through it while still running 64bit encoders.

[ndjamena]

You mean something like this?

avs2pipemod -y4mp %AVSFILE% | "X264.64.exe" %command% --output %VideoFile% --stdin y4m -

[foxyshadis]

Heh, I didn't even realize this was about SD.

Originally Posted by pcordes

Anyway, apparently QTGMC is the best available free (but not open-source?) deinterlacer. And it's still stuck on 32bit only. Next time I have to deinterlace something, maybe I'll find a way to pipe data through it while still running 64bit encoders.

QTGMC and the plugins it uses are all open-source, although there is a closed-source version of mvtools you can substitute in.

For 64-bit, it's also been ported to VapourSynth, a Python-based remake of AviSynth where everything is natively both 32- and 64-bit. Documentation on it hasn't really made it to the avisynth.nl yet.

[ChiDragon]

Originally Posted by pcordes

lossless libx264 (-preset ultrafast -qp 0) will use a lot less disk space, and is quite fast to encode and decode, thanks to ASM-optimized decoders. I'd recommend using it over ffvhuff or utvideo for lossless scratch files

I haven't found this to be the case with YUY2 VHS captures, even with slow encode settings (but I always use --tune fastdecode). Have you seen significant space savings with noisy sources?

[jmac698]

Originally Posted by Ghitulescu
I noticed we were discussing SD, not HD, but with VHS or S-VHS even 8b is too much. Capturing them in 10b or more is a waste of resources.

I just want to dispel this myth. I've personally measured various capture cards and found a bowed curve in the levels which is a typical artefact of a video amplifier. What this means is that a video level of 128 actually read as 131, for example. So you've lost about 1.5 bits around the middle of the luminence scale.

What 10bits is far is basically a more accurate 8 bits, since old 8bits adc's were never actually accurate to 8 bits (called integral non-linearity, you can look up this number on data sheets).

In fact there's fully analog VHS tapes, and if you wanted to record a still frame for several frames you can extract more than 8 bits from it, by simple averaging. The same concept is used in astrophotography and is called stacking. This could be of use to bringing out hidden details in dark recordings for example.

Not going to start an argument here, just some things to think about.