add alias for training models

[allenday]

I see that this repo uses the pre-generated models from the sr repo:

https://github.com/MIR-MU/ffmpeg-tensorflow/blob/master/script/produce-sr-models#L18

Similarly to how you're aliasing the ffmpeg command to transcode
https://github.com/MIR-MU/ffmpeg-tensorflow/edit/master/README.md#L59

It would be great to also add an alias for training custom models, probably requiring some munging of:
https://github.com/HighVoltageRocknRoll/sr/blob/master/generate_datasets.sh
and
https://github.com/HighVoltageRocknRoll/sr/blob/master/train_srcnn.sh

[allenday]

if you can do this I'll generate a model and send PR to this repo if you like @Witiko

[Witiko]

@allenday That's a great idea. We are currently not shipping with Python, only FFMpeg with the Libtensorflow C library. Including Python would make the Docker image considerably larger and most users may not find it useful, so perhaps we could fork HighVoltageRocknRoll/sr and make a separate Docker image for training SRCNN, ESPCN, VSRNet, and VESPCN instead?

[allenday]

sure that works - have two distinct dockerfiles