[FFmpeg-devel] 10-bit DNxHD support

Baptiste Coudurier baptiste.coudurier
Thu Mar 10 01:52:43 CET 2011

On 03/09/2011 06:45 AM, Joseph Artsimovich wrote:
>>> I'd also like to make you aware of av_clip16, you should use that instead of the If (v<0) v=0
>>> ... code you have there.
>> Actually the signal range for HD is 4-1019, written in the specs.
> Yes, that applies to the Y component according to specs.  My code was clipping RGB values to full
> range though, which I believe is a fair deal.

This can produce artifacts if you export to uyvy422 and check in quicktime for example, maybe the 
clipping should be done on the writing side though, not sure.

> Speaking about ranges, VC-3 specs say the range of DCT coefficients is [-1024, 1023] for 8 bit
> and [-4096, 4095] for  10 bit.  However, the values coming from dnxhd_decode_dct_block() are
> sometimes outside of those ranges, taking into account the bias applied to make them
> non-negative.

Somebody more familiar with the idct can probably comment on this.

> I am not sure if it's OK or not, but I did notice ffmpeg doesn't produce
> pixel-to-pixel identical results compared to Avid's decoder.  The difference can most easily be
> noticed on smooth gradients.

Do you have screenshots ? "smooth gradients" is overly subjective.
If you have the vc-3 reference test suite, you can check against these.

In any case it would be nice to have identical output, although you never know with

Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer                                  http://www.ffmpeg.org

More information about the ffmpeg-devel mailing list