[FFmpeg-cvslog] r24134 - in trunk: Makefile tests/fate-run.sh
Måns Rullgård
mans
Fri Jul 9 15:15:49 CEST 2010
Michael Niedermayer <michaelni at gmx.at> writes:
> On Fri, Jul 09, 2010 at 02:33:18AM +0100, M?ns Rullg?rd wrote:
>> Michael Niedermayer <michaelni at gmx.at> writes:
>>
>> > On Fri, Jul 09, 2010 at 02:40:51AM +0200, mru wrote:
>> >> Author: mru
>> >> Date: Fri Jul 9 02:40:51 2010
>> >> New Revision: 24134
>> >>
>> >> Log:
>> >> fate: add oneoff comparison method
>> >>
>> >> This adds the 'oneoff' comparison method. It compares two s16le pcm
>> >> files with a max difference <= the fuzz factor treated as success.
>> >> Default fuzz is 1. On failure, the full tiny_psnr output is printed.
>> >
>> > it also might make sense to bound the sum of squared differences, and not
>> > just the maximum. A file having 1% of samples differing +-1 and one that is
>> > in 50% of samples +-1 off are not the same thing
>>
>> With different floating-point rounding modes, it is perfectly normal
>> to have all negative samples off by one (round towards zero vs -inf).
>> That is the type of test this is intended for.
>
> actually, floating point decoders should output floating point samples
> once all the issues with that have been solved.
The output is still likely to differ a little. When the day comes for
this, we'll have to update tiny_psnr to handle floats too...
Speaking of tiny_psnr, would a patch to make it detect wav headers be
acceptable?
--
M?ns Rullg?rd
mans at mansr.com
More information about the ffmpeg-cvslog
mailing list