[Libva] Will the intel-driver VPP support motion-adaptive deinterlacing on SNB/IVB?

Simon Farnsworth simon.farnsworth at onelan.co.uk
Fri Sep 13 06:22:13 PDT 2013


On Friday 13 September 2013 16:09:04 Xiang, Haihao wrote:
> On Fri, 2013-08-30 at 11:43 +0100, Simon Farnsworth wrote:
> > Hello,
> > 
> >  
> > 
> > We're attempting to use VA-API on Intel Gen 6 and Gen 7 hardware to
> > replace software decoders, and we're hitting quality issues with the
> > VPP deinterlacing support. Our CPUs range from Celeron 847 (low-end
> > SNB) to Core i3-3220 (mid-range IVB).
> 
> The Bob DI quality is improved both for SNB and IVB in master and
> staging branch, could you give a try ?
> 
Using gst-plugins-vaapi, 0.4-branch, and staging branches of libva and
intel-driver, I'm not seeing deinterlacing working at all.

How can I debug this?

For reference, my git state is:

: ~/wc/vaapi/intel-driver (git) [staging]
: sfarnsworth host64  $ git show
commit 8a0a61ce47e086ac156602c3daaf70f20e4ca222
Author: Xiang, Haihao <haihao.xiang at intel.com>
Date:   Fri Sep 13 15:47:50 2013 +0800

    VPP: Remove an assert on forward temporal reference on IVB
    
    Add a warning and return VA_STATUS_ERROR_INVALID_PARAMETER instead
    
    Signed-off-by: Xiang, Haihao <haihao.xiang at intel.com>

: ~/wc/gstreamer/gst-plugins-vaapi (git) [0.4]
: sfarnsworth host64  $ git show
commit 408cb29a7d864eb4c7e9c0ef56ad9e37cd676c29
Author: Wind Yuan <feng.yuan at intel.com>
Date:   Mon Aug 26 13:09:37 2013 +0800

    h264encoder: fix issue of b frame hang
    
     When b frame set in h.264 encoder, need to signal every
     busy buffer to wakeup next thread to catch

: ~/wc/vaapi/libva (git) [staging]
: sfarnsworth host64  $ git show
commit 29ad40444f3ed68e5f54ddd70dd256ea41e0456f
Author: Xiang, Haihao <haihao.xiang at intel.com>
Date:   Tue Sep 10 12:22:39 2013 +0800

    avcenc: support large source file
    
    Signed-off-by: Xiang, Haihao <haihao.xiang at intel.com>


> > 
> >  
> > 
> > Specifically, it appears to only support bob deinterlacing, which on
> > the common "ticker tape" horizontal scrolling text used on news
> > channels looks awful (you can see the alternating fields bob up and
> > down.
> > 
> >  
> > 
> > The description of the deinterlace algorithm in the public SNB
> > documentation, volume 4 part 1 section 2.8, suggests that the hardware
> > is capable of motion adaptive deinterlacing to get better results.
> > Will this be enabled in the intel driver at some point?
> 
> motion adaptive deinterlacing is enabled on IVB too. 
> 
> You can use gst-vaapi to test BobDI/MADI:
> 
> 1> clone 0.4 branch from git://gitorious.org/gst-plugins-vaapi/gst-plugins-vaapi.git
> 2> gst-launch-0.10 filesrc location= <URL> ! ... ! vaapidecode ! vaapipostproc deinterlace=1 deinterlace-method=1 ! vaapisink  # BobDI
>    gst-launch-0.10 filesrc location= <URL> ! ... ! vaapidecode ! vaapipostproc deinterlace=1 deinterlace-method=3 ! vaapisink  # MADI
> 
> > 
> 
> 
-- 
Simon Farnsworth
Software Engineer
ONELAN Ltd
http://www.onelan.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freedesktop.org/archives/libva/attachments/20130913/4cc951e0/attachment.pgp>


More information about the Libva mailing list