To 10bit or not to 10bit!

[liquidskin76]

Hey all,

Just wanted to check opinion on whether 10bit x264 is now the way to go, bearing in mind that the decoder i'll be using (ffmpeg/XBMC) now supports 10bit.

My main goal is to reduce file size compared to 8bit. Is there any reason why i shouldn't crack on and start using 10bit for my encodes from now on?

Encode times are a bit longer, however not too fussed about that.

Cheers

[Atak_Snajpera]

http://x264.nl/x264/10bit_02-ateme-why_does_10bit_save_bandwidth.pdf

You must remember that 10bit won't play on any other device (TV / Media Players / PS3 ...) also DXVA won't work either.
10 bit will work only on HTPC (software decoding)

[liquidskin76]

Hey Atak,

Cheers for the info. Forgot about hardware decoding.

I'll stick with 8bit!

Cheers

[benwaggoner]

Honestly, I think 10 bit is a lost cause for mainstream H.264 use. Higher precision encoding probably will come with H.265, but not before.

[Blue_MiSfit]

I completely agree. AVC Intra / Ultra are the only places we'll see 10 bit H.264 in any sort of widespread use.

[turab]

10-bit encoding is actually quite popular amongst anime fansubbing groups.

[benwaggoner]

10-bit encoding is actually quite popular amongst anime fansubbing groups.
Who have been a quite technically influential group for many years, to be certain, but are hardly mainstream. There are way more iPads in the world than there are anime fansub watchers.

[Audionut]

Some of us don't like conforming to how apple thinks we should do things :)

[ajp_anton]

Unfortunately, we are a small group of people not worth making 10-bit compatible hardware for. We'll just have to wait until 10-bit becomes the next marketing trick for the masses.

[Mug Funky]

you'll need mainstream display devices that can reliably show the difference for a start.

10-bit is good as a mezzanine format, and for some odd corner-cases of the encoding world (anime is but a genre, fansubbers aren't the authority, though they're an interesting case. they also seem to believe that anime is the hardest thing to encode well, though this hasn't been true since the introduction of h.264. banding is only an issue if you insist on scrubbing out not only the noise, but most of the detail as well).

i would LOOOVE it if i could buy a $400 TV that is accurate to 10 bits in all the channels. until then, 10-bit distribution is just one of those OCD settings, like --me tesa, well in the realm of diminishing returns.

[turab]

until then, 10-bit distribution is just one of those OCD settings, like --me tesa, well in the realm of diminishing returns.
You'd be surprised at the amount of compression with certain sources. With film, I've personally noticed a 5-10% reduction in file size, so it's probably not worth it, but with anime, I hear it's more -- maybe 20-30% -- which is quite significant.

[benwaggoner]

10-bit is good as a mezzanine format...
Relatd question: What video tools support decoding 10-bit from 10-bit H.264 bitstreams? Alas, most encoding tools don't even have 10-bit pipelines, so I fear those that can decode 10-bit H.264 might wind up truncating to 8-bit and going from there.

But yes, it would be a GREAT mezzanine alternative to the industry standard MPEG-2 (inefficient, generally less than pristine quality) and ProRes (HUGE!).

[benwaggoner]

Also, has x264 or other H.264 encoders looked at retaining all 10 bits of 10-bit sources to the quantization stage, even when doing an 8-bit encode? It seems that having more precise data would enable the encoder to better determine when to dither and when not to, gaining some of the value of 10-bit encoding.

Essentially, a perceptually optimized in-loop dither.

Now that the 10-bit source piepline exists for x264...

[RunningSkittle]

x264 is 16bit internally.
http://forum.doom9.org/showpost.php?p=1415170&postcount=24

[benwaggoner]

x264 is 16bit internally.
http://forum.doom9.org/showpost.php?p=1415170&postcount=24
Cool.

But, by default, are 10-bit sources input at 10-bit into the 16-bit mode? If input is YV12, it's still only 8-bit.

[mandarinka]

x264 is 16bit internally.
http://forum.doom9.org/showpost.php?p=1415170&postcount=24

You are misinterpreting that. What he means there is that doing the math involved requires 16-bit integers just to store *10-bit* values and perform operations on them.
Similarly, with 12bit bitdepth, IIRC some operations would even require moving to 32-bit integers, because 16bit ones wouldn't have enough headroom.

As you can read there though, the values need to actually be clipped to stay within the bitdepth. So saying that x264 is "internally working in 16bit" is incorrect.

[MasterNobody]

x264's filter chain works at 8 or 16-bit depth depending from input and filters. Before sending it to libx264 (actual encoding) it is converted to 8 or 10-bit according to build version (downsampling is made with dithering). So no, actual dithering is made before encoding, not in-loop.

[benwaggoner]

So, what I'd love to see is 10-bit input into quantization, and the final dithering decision being made at that stage based on what will look and compress the best.

[kieranrk]

So, what I'd love to see is 10-bit input into quantization, and the final dithering decision being made at that stage based on what will look and compress the best.

It was discussed briefly at VideoLan Developer Days last year but that's about it.

[Sapo84]

they also seem to believe that anime is the hardest thing to encode well, though this hasn't been true since the introduction of h.264. banding is only an issue if you insist on scrubbing out not only the noise, but most of the detail as well.

Interesting point of view given the fact that banding is present in (many) Cruncyroll encodes, (many) transport stream and (lots of) blu-ray (madoka and the -monogatari series are a fine example of that).
Sometimes there is banding even in the master itself (madoka, confirmed by mp3dom).

That means fansubbers have to deband even if they would prefer not to (already grainy BD source), and 10bit helps a lot in preserving the gradients without going insane with debanding/addgrain.
It's not that anime are difficult to encode, the problems lies in the fact that they are produced by incompetent people.

And that's why the good ones started using 10bit, the vast majority of clueless rippers/encoders just followed because 10bit was the next cool thing.