[FFmpeg-devel] [GSOC] [PATCH] DNN module introduction and SRCNN filter update

[Sergey Lavrushkin]

2018-05-24 22:52 GMT+03:00 James Almer <jamrial at gmail.com>:

> On 5/24/2018 4:24 PM, Sergey Lavrushkin wrote:
> > Hello,
> >
> > This patch introduces DNN inference interface and simple native backend.
> > For now implemented backend supports only convolutions with relu
> activation
> > function, that are sufficient for simple convolutional networks,
> > particularly SRCNN.
> > SRCNN filter was updated using implemented DNN inference interface and
> > native backend.
> >
> >
> > adds_dnn_srcnn.patch
> >
> >
> > From 60247d3deca3c822da0ef8d7390cda08db958830 Mon Sep 17 00:00:00 2001
> > From: Sergey Lavrushkin <dualfal at gmail.com>
> > Date: Thu, 24 May 2018 22:05:54 +0300
> > Subject: [PATCH] Adds dnn inference module for simple convolutional
> networks.
> >  Reimplements srcnn filter based on it.
> >
> > ---
> >  Changelog                      |   2 +
> >  libavfilter/vf_srcnn.c         | 300 +++------------
> >  libavfilter/vf_srcnn.h         | 855 ------------------------------
> -----------
> >  libavutil/Makefile             |   3 +
> >  libavutil/dnn_backend_native.c | 382 ++++++++++++++++++
> >  libavutil/dnn_backend_native.h |  40 ++
> >  libavutil/dnn_interface.c      |  48 +++
> >  libavutil/dnn_interface.h      |  64 +++
> >  libavutil/dnn_srcnn.h          | 854 ++++++++++++++++++++++++++++++
> ++++++++++
> >  9 files changed, 1455 insertions(+), 1093 deletions(-)
> >  delete mode 100644 libavfilter/vf_srcnn.h
> >  create mode 100644 libavutil/dnn_backend_native.c
> >  create mode 100644 libavutil/dnn_backend_native.h
> >  create mode 100644 libavutil/dnn_interface.c
> >  create mode 100644 libavutil/dnn_interface.h
> >  create mode 100644 libavutil/dnn_srcnn.h
>
> With this change you're trying to use libavformat API from libavutil,
> which is not ok as the latter must not depend on the former at all. So
> if anything, this belongs in libavformat.
>
> That aside, you're using the ff_ prefix on an installed header, which is
> unusable from outside the library that contains the symbol, and the
> structs are not using the AV* namespace either.
> Does this need to be public API to begin with? If it's only going to be
> used by one or more filters, then it might as well remain as an internal
> module in libavfilter.
>

Yes, I think it will be used only in libavfilter. I'll move it there then.
And should I use  ff_ prefix  and  AV* namespace for structs in
an internal module in libavfilter?


> And you need to indent and prettify the tables a bit.
>

Do you mean kernels and biases in dnn_srcnn.h?
Their formatting represents their 4D structure a little bit and it is
similar to one that I used for the first srcnn filter version that was
successfully pushed before.