Re: [Libquicktime-devel] (no subject) <- Please add one next time :)

[Burkhard P. <bur...@ig...>]

Am 28.11.2013 16:06, schrieb Erik Johansson:
> Two patches for color model handling: The first adds support for
> conversions from BC_YUVJ420P to BC_YUVJ422P.

Applied

> The second changes the
> way the function get_conversion_price() calculates the price.
>
> The initial issue I had with the latter was that when converting for
> example YUV420P to either YUV422P or YUV444P they were assigned the
> same score.  I would prefer a preference for 422 over 444 because that
> adds less extra information in the conversion.

Good to see that someone works on that code :)
Some remarks:

- colormodel_is_legal_range() is a misleading name. Are there illegal ranges too?
   I'd call it lqt_colormodel_is_video_range(). Maybe it can even be made public.

- The approach for calculating the price for different chroma subsampling is
   much better, but not optimal yet. With your approach, converting 4:4:4 to 4:2:2
   has the same price as 4:4:4 to 4:1:1. Could we think about something like:

   if(input_sub_h < output_sub_h)
     price += some_constant * (output_sub_h - input_sub_h);

- The subsampling should also take RGB formats into account (with sub_h = 1 and
   sub_v = 1). BC_RGB888 -> BC_YUV420P should be worse than BC_RGB888 -> BC_YUV444P,
   or not?

Burkhard