[FFmpeg-devel] [PATCH 1/2] dnn: add layer pad which is equivalent to tf.pad

[Guo, Yejun]

> > -----Original Message-----
> > From: Guo, Yejun
> > Sent: Monday, July 01, 2019 4:08 PM
> > To: ffmpeg-devel at ffmpeg.org
> > Cc: Guo, Yejun <yejun.guo at intel.com>
> > Subject: [PATCH 1/2] dnn: add layer pad which is equivalent to tf.pad
> >
> > the reason to add this layer first is that vf_sr uses it in its
> > tensorflow model, and the next plan is to update the python script
> > to convert tf.pad into native model.
> >
> > Signed-off-by: Guo, Yejun <yejun.guo at intel.com>
> 
> this patch set is based on a previous patch at
> http://ffmpeg.org/pipermail/ffmpeg-devel/2019-June/245481.html
> (libavfilter/dnn: move dnn files from	libavfilter to libavfilter/dnn)

hello, I've finished 10+ new patches locally to improve dnn module, including one new patch relative to pad.

I'll resend this patch set, together with the new pad relative patch.

I'll send more patches set by set, once the previous patch set is pushed, since the patches have dependency.

Just in case you are interested in my new patches, I've uploaded to https://github.com/guoyejun/ffmpeg/tree/dnn0716. 
for your convenient, I also copy the oneline log here for each patch (from newer to older) with 3 patch sets.

50a3353 fate: add unit test for dnn depth_to_space layer
af9e3ab dnn: separate depth_to_space layer from dnn_backend_native.c to a new file
41b97e4 fate: add unit test for dnn conv2d layer
4143485 dnn: separate conv2d layer from dnn_backend_native.c to a new file

870383e dnn: export operand info in python script and load in c code
650d576 dnn: change .model file format to put layer number at the end of file
d029bf8 dnn: introduce dnn operand (in c code) to hold operand infos within network

c9b9e1c doc/filters: update how to generate native model for derain filter
064aa45 convert_from_tensorflow.py: support conv2d with dilation
1c419a5 convert_from_tensorflow.py: add option to dump graph for visualization in tensorboard


I'll continue to improve dnn module, such as:
- update dnn interface since we have more operand info (names, types, dims) in native mode.
- optimize native conv2d layer
- add more native layers to support at least mobile net as a milestone
- add a general filter which generates analysis result (in a general side data for AVFrame) with dnn network.
- add a general filter which changes the content of AVFrame with dnn network.