[FFmpeg-devel] [PATCH] dnn_backend_tf.c: add option sess_config for tf backend

[Guo, Yejun]

> From: Chris Miceli <chris at miceli.net.au>
> Sent: 2020年10月14日 6:15
> To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> Cc: Guo, Yejun <yejun.guo at intel.com>
> Subject: Re: [FFmpeg-devel] [PATCH] dnn_backend_tf.c: add option sess_config
> for tf backend
> 
> Hi!
> 
> With proto files they can be serialised to many different formats, is there a
> reason for it to use hex? The reason I ask is that it's not trivially readable for
> those debugging the command line. Perhaps something like json is better?

yes, readable message as FFmpeg command line option is very good, but the readable
message need to be serialized into proto within ffmpeg since tensorflow C api only accepts
the serialized proto, it means that we need to:
- add libprotobuf dependency to ffmpeg
- know tensorflow's model format so to generate serialized proto.

these are not feasible, so ffmpeg has to accept the serialized proto which need to be
generated out of ffmpeg.

> 
> Even if not using json, maybe a file is a little easier? In cases where people have a
> large system using this, it just seems like it would be easier to chase down issues.

yes, serialized proto can be in a file, we can add such option sess_config_file once a
customer requires it. As the first step, we can use the command line directly. thanks.

> 
> Thanks,
> Chris