Re: getframe() alternatives

[Ben Abbott]

On Oct 5, 2011, at 9:07 AM, PhilipNienhuis wrote:

> While porting a ML script that creates a video clip to Octave I stumbled upon
> the lack of getframe().
> The OF video package (or should it be: the image package) still lacks that.
> 
> The reason I ask on Maintainers is that I seek for coding advice.
> 
> For now I have the following clumsy solution for adding successive screen
> images (plots) to an avi stream:
> 
> Alternative 1
> ==========
> <for ...>
>  <...create plot with filled contours, pimp it...>
>  print ("tmp.fig", "-dppm");
>  HH = im2double (imread ("tmp.fig"));
>  aviobj = addframe (aviobj, HH);
> <endfor>
> 
> ...which works. But of course, very slowly. 
> (FYI: im2double() (OF image pkg) converts to double and scales to [0..1]
> just as addframe() (OF video pkg) requires.)
> 
> 
> Alternative 2
> ==========
> I figured (but I may be wrong) that skipping the disk write/read portions
> may save considerable amounts of time. So would it be possible to have
> print() write to a pipe rather than file, and imread() read from a pipe
> rather than file? That might be the most platform-independent "kludge".
> 
> 
> Alternative 3
> ==========
> Obviously a genuine binary getframe() solution is preferable. Apart from
> skipping disk I/O it would avoid double graphics format conversion as well
> and simply transfer the screen image representation in video memory into a
> bitmap in the form of a 3D array.
> 
> On Windows (where I work on for this ML script) this should be a breeze once
> the screen API graphics handle is available. FLTK would be the way to go
> then; Gnuplot is a separate prog so access to its internals is probably a
> bit hard. 
> Over a decade ago I once made something like that for a screen grab routine
> in assembly language; however yesterday a google search turned up dozens of
> much better designed examples in C / C++ / C# etc.
> However I have no idea where to search for the graphics handle in the
> Octave/FLTK sources. Or is it hidden in the FLTK libs and never used
> directly by Octave?
> 
> On Linux I wouldn't know how to proceed, but I presume graphics handles are
> also used there.
> 
> Is there any advice on how to most efficiently get a bitmap/3D array out of
> an Octave graphics handle? perhaps even code snippets?
> 
> Thanks,
> 
> Philip

I expect the fastest approach would be to capture the image data directly from 
OpenGL. That would create mean the gnuplot backend wouldn't work for 
getframe(), but I think that is ok in the long run.

Unfortunately, I don't know enough about the OpenGL internals to be of any help 
in how to grab the image data. But I did dig up the link below which Jordi 
posted to the list in 2010.

        http://www.opengl.org/resources/faq/technical/miscellaneous.htm#misc0050

Looks like a good place to start?

Ben