[FFmpeg-devel] VC-1 Interlaced Hardware decoding

Jeroen de Kleijn jeroen.de.kleijn at gmail.com
Sat Nov 17 12:13:51 CET 2012


Hello Hendrik,

I have been working on getting your patches working in OpenElec, which uses
XVBA gpu acceleration directly and not via VAAPI under Linux. Without your
patches only the I-frames and P-frames were shown, since the
B-frame code-path was missing. With your patches I got the B-frames
working. Thank you very much for that.

However, while testing with the VC-1 1080i BBC planet earth sources I still
see blocking artifacts occasionally. These were already showing before I
merged the patches and thus seem unrelated to the B-frame code-path
missing. I was wondering if you can test the sample mentioned below with
your build using DXVA2 acceleration to see if the artifacts also show up?
This could give some hints if this is still a FFMpeg issue or an issue with
how XVBA acceleration is used. The artifacts show up at scene changes, e.g.
at 52 seconds in the sample and couple of time more later on in the sample.
It seems like if not the last reference frame from after the scene change
is used, but the one before the scene change is used for the next couple of
P and B-frames. It is as if the last I-frame was interpreted as a BI-frame.
However, I only see this problem for interlaced content and not for
progressive content.

I've been unable to find the root cause and no issues show up in the
logging. Hopefully you or some other person on the mailing list can
reproduce the issue and have a clue of what is going on and where in the
code.

Greetings Jeroen

sample:
https://www.dropbox.com/s/1te2zll4exg46q7/Sample%20-%20From%20Pole%20to%20Pole%20-%201080i.BluRay.DTSHD-HR.VC-1.mkv


More information about the ffmpeg-devel mailing list