[FFmpeg-devel] H.261 chroma motion vector rounding.

Hendrik Leppkes h.leppkes at gmail.com
Mon Apr 3 00:03:19 EEST 2017

On Sun, Apr 2, 2017 at 10:33 PM, Christina Brien
<christina.brien at polygonindustrial.com> wrote:
> Hi
> Having written a clean-room implementation of H.261 and comparing it to the output of FFMPEG, I have found that FFMPEG is not spec-compliant with regard to negative chroma motion vectors. The spec states that chroma motion vectors are the luma motion vectors divided by two, and rounded *towards zero*. FFMPEG currently rounds towards negative infinity.
> This patch makes FFMPEG compliant, but breaks the regression tests.

C integer divison truncates, that equals rounding towards zero.

- Hendrik

More information about the ffmpeg-devel mailing list