[FFmpeg-devel] [FFmpeg-cvslog] Adds ESPCN super resolution filter merged with SRCNN filter.

[Pedro Arthur]

2018-07-02 14:55 GMT-03:00 Jean-Baptiste Kempf <jb at videolan.org>:
> Hello,
>
> On Mon, 2 Jul 2018, at 19:37, Pedro Arthur wrote:
>> 2018-07-02 14:23 GMT-03:00 Jean-Baptiste Kempf <jb at videolan.org>:
>> > On Mon, 2 Jul 2018, at 19:12, Pedro Arthur wrote:
>> >> 2018-07-02 14:02 GMT-03:00 Jean-Baptiste Kempf <jb at videolan.org>:
>> >> > On Mon, 2 Jul 2018, at 18:46, Sergey Lavrushkin wrote:
>> >> >> ffmpeg | branch: master | Sergey Lavrushkin <dualfal at gmail.com> | Thu
>> >> >> Jun 14 00:37:12 2018 +0300| [575b7189908e1cfa55104b0d2c7c9f6ea30ca2dc] |
>> >> >> committer: Pedro Arthur
>> >> >>
>> >> >> Adds ESPCN super resolution filter merged with SRCNN filter.
>> >> >
>> >> > Where is the source for all those numbers?
>> >> You mean the cnn weights? If yes, the srcnn weights are from the
>> >> original paper implementation in matlab.
>> >
>> > Where do they come from, how can we recreate them?
>> Paper link [1], and web page with reference matlab code [2].
>
> This code is not open source, and is not compatible with LGPLv2.1:
>
> "If you use/adapt our code in your work (either as a stand-alone tool or as a component of any algorithm),
> you need to appropriately cite our ECCV 2014 paper or arXiv paper."
>
> Reimplementation of the code in a different language does not remove IP.
The paper is properly cited in libavfilter/vf_sr.c.

I not specialist in  IP rights but  there is nothing novel in the code
(TBH there is no code at all, look at srcnn.m) it is just plain
convolution, basic math machinery, we are just applying a bunch of
filter convolution to an image.
If you're referring to using his trained data, Ok, we could train our
own weights if it is really an issue.


>
>> >> The espcn was trained by Sergey, if needed we can provide the TF
>> >> model/ training data.
>> >
>> > Same remark.
>> Paper [3], for this model Sergey used his own trained weights.
>> If needed, you can ask him directly for the training data set and
>> TensorFlow model for both methods.
>
> Where is the data set then? How was it trained? How can we reproduce them?
CC'ing Sergey, so he can provide any data you want.

>
> --
> Jean-Baptiste Kempf -  President
> +33 672 704 734
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel