Re: [Openexr-devel] packing rgba into 32 bit int.

[Florian Kainz]

Eric Soulvie wrote:
> We're mainly using the tiled format to dynamically load chunks of large 
> images at render time.  With RIP maps, these files can easily exceed 128 
> megs using the HALF type.  For many of these image the range is not 
> important.   I was wondering if anyone has anyone tried packing RGBA 
> into an int  and using a single channel?
>  
>     Eric
>     Blizzard Cinematics
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Openexr-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/openexr-devel


Instead of packing RGBA data into an int, I would recommend quantizing
the 16-bit data so that only 256 out of the 65,536 are used.  You can
do this by converting your image to 8 bits per pixel and then back to
16 bits.  The quantizing operation can be implemented efficiently using class 
halfFunction.  The resulting OpenEXR file should compress to approximately the 
same size as if you stored actual 8-bit data. (At ILM, we tested this a while ago.)

If you pack RGBA data into an int, tools that filter and resize images, for 
example, exrmaketiled, will not work very well because they won't recognize the 
integer data as packed 8-bit samples.

Is there something special about the 128 meg size?

Florian