[mythtv] [mythtv-commits] Ticket #2649: Opengl video renderer

Steven Adeff adeffs.mythtv at gmail.com
Thu Jan 25 18:18:30 UTC 2007


On 1/24/07, Mark Kendall <mark.kendall at gmail.com> wrote:
> > What do you mean by "progressive scan versions of onefield, linearblend and
> >  kerneldeint"?
>
> (My apologies if this sounds like an egg sucking lesson)
>
> They are frame rate doubling deinterlacers, taking each field in turn
> and using different methods to guess the data in the missing rows to
> produce a complete frame. If there is no motion detected, they just
> use the data from the other field (i.e. 'perfect' quality). Otherwise,
> linearblend interpolates the missing data from the rows above and
> below, kerneldeint takes a view based on certain pixels above, below
> and in the next and previous frames and onefield just uses the data
> from the current field.
>
> The latter is effectively an 'intelligent' bobdeint - it only 'bobs'
> when motion is detected rather than applying it to the entire picture.
> So no more flickering text etc.
>
> Personally, I think the kerneldeint version gives the best results but
> it can definitely be improved upon with better motion detection and
> using edge detection/motion compensation techniques.

ok, so basically what was there before? I guess I was confused by what
"progressive scan versions" meant.

> > What are your prospects in fixing the top-field deitnerlacing issue?
>
> Easy enough - do you have some bottom field first samples that I can test with?

not off hand, I may be able to find some if you need test samples?

> > Does this include a higher quality resize like has been discussed on
> > this list before? (ie SD recordings and DVD playback resizing to 720p
> > or 1080p resolution with all the filter goodies you seem to be
> > implementing)
>
> That's the plan. I'm just not sure how realistic it is yet. Assuming
> quality is your objective, you'll want to use the best progressive,
> deinterlacer (GPU demanding and double frame rate), followed by a
> scaling operation (haven't figured out yet whether this is a simple
> linear/bilinear scale or more complex), followed by applying some sort
> of convolution kernel on your now much larger image. That last stage
> is going to be really hard on your graphics hardware - to the extent
> I'm even wondering whether it will run in realtime on reasonable
> hardware.

Well, I hope for DVD sized video's it can do it. I know in mplayer the
difference between resizing a DVD to 1080p using its lanczos filter is
not very taxing on my Athlon64 3200+ so I'm hoping it will be
possible.


> > I've got a 6200TC on my dedicated frontend, which currently runs
> > 0.20-fixes, is your work dependent on being compiled with SVN?
>
> The patches are against the mythtv-vid branch.

good to know, thanks!

-- 
Steve
Before you ask, read the FAQ!
http://www.mythtv.org/wiki/index.php/Frequently_Asked_Questions
then search the Wiki, and this list,
http://www.gossamer-threads.com/lists/mythtv/
Mailinglist etiquette -
http://www.mythtv.org/wiki/index.php/Mailing_List_etiquette


More information about the mythtv-dev mailing list