Hi

Just for reference, looks like the gstreamer community is circling
around reusable colorspace conversion plugins as well as optimized
glsink (recalling the quite recent media-center-related threads on
this very ml).

How shall such improvements be integrated in clutter's gst video sink
? Using it in place of ffmpegcolorspace elements, or should we
directly improve the cluttergst module ?

Are there perspectives on using a "stock/legacy" glimagesink as
cluttergst videosink (through context sharing?)?

For benchmarking purposes, i did some preliminar tests about clutter
VS pigment, which seems to perform (thanks to accelerated colorspace
conversion?) around 1.6 times better (based on CPU load, test case:
httpsrc ! jpegdec ! ffmpegcolorspace ! videosink, VGA resol @ 30 FPS).
Looks like there's space for video optimization...

Regards,

Florent

---------- Forwarded message ----------
From: Richard Spindler <[EMAIL PROTECTED]>
Date: Mon, May 19, 2008 at 12:45 PM
Subject: Re: [gst-devel] gst-plugins-gl, OpenGL and stuff
Cc: [EMAIL PROTECTED]


Hi,

2008/5/19 KwangYul Seo <[EMAIL PROTECTED]>:
> Do you have any progress in forming a Linux+Video+OpenGL working group?

Not yet, but I was made aware of a very cool implementation of OpenGL
Color-Space Conversion at

http://yuvtools.wiki.sourceforge.net/

The code is very readable and flexible in regard to different color-formats.

Unfortunately I head some difficulties running this code on my ATI
hardware, due to issues with the handling of non-power-of-2 textures
on ATI Hardware. It works for some video sizes which happen to fit
into that limitation. I think it should work better on nvidia
hardware.

Other than that, I found that the "FreeFrame" Video Filter Plugin
Specification http://freeframe.sf.net/ specifies reusable OpenGL based
image processing algorithms. Eventually something similar should be
possible for "pluggable" colorspace conversion.

Currently I am using the "gavl" library for software colorspace
conversion, which has a rich collection of low-level convertors, and
could serve as a good fallback option for low-end opengl hardware.
This library is a sub-project from http://gmerlin.sf.net/

> Currently I am looking for a colorspace conversion solution implemented
> using OpenGL. It is not hard to find one, but most solutions use high-end
> video card extensions and do not provide any software fallback.

I've been playing around with creating some kind of Abstraction-Layer
for the code from yuvtools, such that it can be easily replaced by an
alternative implementation, but so far I am not quite satisfied with
my work, so I have not yet published anything.

> I think we must build reusable software to do low level tasks such as image
> composition, video mixsing and colorspace conversion which are repeated over
> and over again in constructing various video applications.

I totally agree, and in my opinion the most promising project in that
respect is gmerlin/gavl. I've been in close contact with its lead
developer, and I think it is really great. It is more "low-level" then
gstreamer, but for some types of applications this is actually an
advantage.

Cheers
-Richard
-- 
To unsubscribe send a mail to [EMAIL PROTECTED]

Reply via email to