[FFmpeg-devel] Interpolation algorithms for FFmpeg

Loren Merritt lorenm
Tue Aug 3 15:27:37 CEST 2010


On Tue, 3 Aug 2010, Sebastian Vater wrote:
> Loren Merritt a ?crit :
>> On Wed, 28 Jul 2010, Jason Garrett-Glaser wrote:
>>> On Wed, Jul 28, 2010 at 6:06 PM, Michael Niedermayer
>>>
>>> Because it's written purely in float and unnecessarily slow, really.
>>> IIRC, Loren suggested it could be at least 10x faster with hardly any
>>> changes to the overall concept, which is enough to at least do SD in
>>> realtime.
>>
>> Tritical has fixed some parts since then, so I'd estimate 3x now.
>
> Maybe that would be a good task then for bringing that 10x faster in
> total for the Bachelor work as well as for FFmpeg?
>
> If I'm not wrong then most of the speed gain can be achieved by doing a
> 100% integer (fixed point) version?

float->int16 approximately doubles throughput. Or converting only most of 
it to int16 (which I have done) is 1.8x so far (over nnedi3-0.9.1). int8 
would be faster, but also loses a signficant amount of precision, so I 
ruled that out.
The remaining optimization idea I had involves exploiting isotropy to 
reduce the number of edge orientations the ANN has to deal with. But that 
involves retraining, so it'll take longer to test.

--Loren Merritt



More information about the ffmpeg-devel mailing list