x264 10bit-depth

[Anakunda]

hello,

today a 10 bit backend was presented. I wanted to ask if I can replace this backend with current x264, and furthermore if 10bit backend adds some quality to video, or will the video be bigger only without additional quality?
Mostly I encode from DVD and DVB(MPEG2).

[Lyris]

Those are both 8bit sources anyway.

[Blue_MiSfit]

Hi Anakunda,

10 bit is a tricky thing. Here are the main talking points currently, as I see them: (Take these with a grain of salt, as I'm not terribly knowledgable or experienced WRT 10 bit x264!)

1) 10 bit H.264 can only be decoded by payware decoders currently, like Mainconcept.
2) 10 bit encoding is best used with 10 bit sources, of which there are certainly none available to the average consumer
3) 10 bit gives better quality at the same bitrate AFAIK, due to greater precision
4) 10 bit is extremely slow currently, because ASM optimizations haven't been committed to the GIT yet. This WILL happen eventually :)

So, my advice would be to skip it for now. This will certainly change in the future!!

Derek

[Astrophizz]

I would think that 10 bit would also be good for 8 bit sources since it is lossy encoding and the increased precision would help.

[burfadel]

What does the 10-bit precision mean anyway? Is it possible to have the encoder chain 10-bit then have it downsample to 8-bit precision for the final encode?

[aegisofrime]

Hi Anakunda,

10 bit is a tricky thing. Here are the main talking points currently, as I see them: (Take these with a grain of salt, as I'm not terribly knowledgable or experienced WRT 10 bit x264!)

1) 10 bit H.264 can only be decoded by payware decoders currently, like Mainconcept.
2) 10 bit encoding is best used with 10 bit sources, of which there are certainly none available to the average consumer
3) 10 bit gives better quality at the same bitrate AFAIK, due to greater precision
4) 10 bit is extremely slow currently, because ASM optimizations haven't been committed to the GIT yet. This WILL happen eventually :)

So, my advice would be to skip it for now. This will certainly change in the future!!

Derek

Thanks so much for summarizing it so clearly and succinctly :) It really clears up the confusion I had.

[Dark Shikari]

2) 10 bit encoding is best used with 10 bit sources, of which there are certainly none available to the average consumerIt works just fine with 8-bit too. We will be adding things like gradfun into the filter chain to allow 8-bit stuff to be automatically upsampled to 10-bit anyways.

Of course, widespread adoption will depend on libavcodec getting 10-bit support.

[Selur]

We will be adding things like gradfun into the filter chain to allow 8-bit stuff to be automatically upsampled to 10-bit anyways.
Nice!

[J_Darnley]

What does the 10-bit precision mean anyway? Is it possible to have the encoder chain 10-bit then have it downsample to 8-bit precision for the final encode?

No.
5

[Amefurashi]

Of course, widespread adoption will depend on libavcodec getting 10-bit support.

Is there the technical possibility (as far as you would know) to implement the 10-bit decoding in common standalone devices as well (eg. WDTV Live, Popcorn Hour etc.)?

[Dark Shikari]

Is there the technical possibility (as far as you would know) to implement the 10-bit decoding in common standalone devices as well (eg. WDTV Live, Popcorn Hour etc.)?Sure, but it would probably require new hardware decoders.

[Amefurashi]

Sure, but it would probably require new hardware decoders.

Ouch... I was hoping for a firmware-updating solution...

Thanks :)

[burfadel]

No.
5

I didn't think so :)

[lithoc]

Sure, but it would probably require new hardware decoders.

How about UVD or Purevideo/VDPAU or intel?
May be we will have another round of ... "Who will be the 1st to support High10P"?

It's good to see some competition here.
:)

[LoRd_MuldeR]

How about UVD or Purevideo/VDPAU or intel?

As the do not use GPU programs (shaders), which probably could be updated, but "hardwired" decoder chips integrated on the graphics board, which can not be updated, new hardware will be nedded :rolleyes:

[Sharktooth]

..why not add gradfun (in addition to the standard deblock) into the decoder specs? it would save a lot of headaches...

[UltraTV]

What colorspace should be used for 10-bit encoding in x264? I have some 4:2:2 sources, but when I run them through they're being converted to yuv420.

x264_10 --pass 1 --preset veryslow --tune film --profile high10 --bitrate 30000 --stats ".stats" --keyint 96 --thread-input --sar 1:1 --output NUL "test.avs"avs
[info]: 1920x1080p 1:1 @ 29970000/1250071 fps (cfr)
resize [warning]: converting from yuyv422 to yuv422p
resize [warning]: converting from yuv422p to yuv420p.........

[J_Darnley]

x264 only supports 4:2:0

[UltraTV]

I see, so 4:2:0 @ 10-bit. So sticking with yv12 is best.

[popper]

Sure, but it would probably require new hardware decoders.

How about UVD or Purevideo/VDPAU or intel?
May be we will have another round of ... "Who will be the 1st to support High10P"?

It's good to see some competition here.
:)

actually DS perhaps you might ask Francois about the SB Decoder ?, i think it will/might, as its not a fixed ASIC fixed-function decoding device such as NV and amd seem to use now, and he could probably work it into the patch if not already there somehow if it ever goes through.

[Dark Shikari]

actually DS perhaps you might ask Francois about the SB Decoder ?, i think it will/might, as its not a fixed ASIC fixed-function decoding device such as NV and amd seem to use now, and he could probably work it into the patch if not already there somehow if it ever goes through.I suspect the SB is more like the most recent iteration of Broadcom's VideoCore; it's not a "pure ASIC", but rather built out of programmable blocks. So, for example, for a motion search, it could load some region of pixels and perform arbitrary checks against it. But that region of pixels still likely has to be 8-bit.

[lovelove]

I come from image editing and I was wondering why did you go from 8 to 10 bit instead of directly to 16 bit?

As far as (still) images are concerned a higher bit depth (16 [or 10] bit compared to 8 bit) is ONLY needed IF 8 bit would result in banding/posterization artefacts. However the price is a bigger file size.

So how do you guys end up saying "10 bit gives better quality at the same bitrate due to greater precision" ??
IMO, the opposite is the case.
10 bit gives marginally better quality (probably only distinguishable for gradients with banding) BUT needs a higher bitrate (due to greater precision).
At the SAME bitrate, I suppose 8 bit would still have better quality.

[Dark Shikari]

I come from image editing and I was wondering why did you go from 8 to 10 bit instead of directly to 16 bit?Because each extra bit of intermediate precision cuts 50% off the amount of compression lost due to intermediate rounding. After 2 extra bits, you've eliminated 75% of the inefficiency.

Also, 10-bit is the highest bit depth at which luma motion compensation can be implemented without unpacking to 32-bit.

Also, the spec is capped at 14-bit.

[dukey]

What's the point of 10bit video ? Most LCD displays only use 6bit per channel and dither the result. Seems like massive over kill to me. In direct x or opengl, you can render to 16 or 32bit per channel buffers, but the final result is always clamped down for display.

[Dark Shikari]

What's the point of 10bit video ? Most LCD displays only use 6bit per channel and dither the result. Seems like massive over kill to me. In direct x or opengl, you can render to 16 or 32bit per channel buffers, but the final result is always clamped down for display.10-bit is the intermediate precision. The output precision isn't important.

Audio has used 32-bit intermediate precision (in the form of float) for decades even though the output is only 16-bit.

[me7]

The final version of Libav 0.7 was released. I guess widespread support of 10-bit AVC is underway.

[CruNcher]

What's the point of 10bit video ? Most LCD displays only use 6bit per channel and dither the result. Seems like massive over kill to me. In direct x or opengl, you can render to 16 or 32bit per channel buffers, but the final result is always clamped down for display.

You see the Industry has something "new" to bring in for Consumers after the 3D Hype (Ultra Precission Display) UPD for short ;)