[FFmpeg-devel] Pixel format for OpenEXR support

[Jimmy Christensen]

On 2009-06-17 22:17, Brian Crowell wrote:
> On Wed, Jun 17, 2009 at 3:07 PM, Michael Niedermayer<michaelni at gmx.at>  wrote:
>> this depend on what they are semantically and what they are used for
>
> It's an intermediate format for film and digital effects work in
> linear RGB, linear YCbCr, or linear gray (with additional channels
> possible).
>
> With integer channels you might have a chance, but you can't convert
> from floating-point to 16 bpc integer without clamping the input
> somewhere, and the standard doesn't specify any reference points.
>

It seems that 0 to 1 float is the norm for clamping to the visual space. 
I've made a quick EXR reader which only reads 32-bit which clamps 0 to 1 
into a rgb48le format. Will add support for the other types and post 
here on the list. Should be fairly easy to modify it later to support 
eg. "rgb96f" and "rgb48f" later. Then the clamping can be done in the 
swscaler.