[FFmpeg-devel] [PATCH] libavfilter: temporarily remove DNN framework and vf_sr filter

Pedro Arthur bygrandao at gmail.com
Fri Jul 27 06:49:41 EEST 2018


2018-07-26 23:19 GMT-03:00 Ronald S. Bultje <rsbultje at gmail.com>:
> Hi,
>
> On Thu, Jul 26, 2018 at 10:04 PM, Pedro Arthur <bygrandao at gmail.com> wrote:
>
>> If you compare NN weights with quantization tables they are pretty
>> similar
>
>
> https://chromium.googlesource.com/webm/libvpx/+/3b9c19aaa7b8830a896c5f578a3ce6c6a7953947%5E%21/#F0
>
> So, that one tiny single function is how VP9/AV1 quant tables are generated.
>
> Or, the HEVC/H264 ones, they are even simpler: exp2(qp/6).
>
> Are NN weights a single, one-line (10-character) expression? Please
> elaborate. Why isn't that 10-character function documented anywhere?
I think you missed the point, I wrote "can be obtained from a training
process over a dataset so it achieves better results
(quality/compression)".
Taking the vp9 as example, sure the coeficients are obtained by the
'poly3' but the real data are the polynomial coeficients, does any one
asks where these polynomial coeficients came from, reproducibility,
etc? Your comparison does not seems fair to me.

>
> Ronald
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel


More information about the ffmpeg-devel mailing list