fixes

[johndpope]

No description provided.

[gael-vanderlee]

Looks good! Could you integrate the docker commands better in the readme ? Perhaps make a "Docker" section in "Getting started" with instructions.

[johndpope]

sure. been battling getting the generated files off the docker image. Is it a case of when the style is trained - a new image can be hotswapped in? or does the new image need training as well? I'm imaging a use case for video - but seems prohibitively (ocmputationally) expensive.

[gael-vanderlee]

Every new image needs training, so this project doesn't work on video (or only very slowly). For video, you could implement existing solutions, there are video segmentation models (such as https://github.com/kmaninis/OSVOS-PyTorch) as well as video style transfer models (such as https://github.com/manuelruder/fast-artistic-videos) that encode temporal information.