Port to replicate (cog.yaml / predict.py)

[platform-kit]

Hi there! Excellent work with this VC model.

I was wondering if you might be willing to write a quick port to support deployment on replicate, which uses Cog to deploy ML models inside Docker containers. Implementation consists of 2 files, a cog.yaml in which you define dependencies and a predict.py in which the relevant inputs are defined and inference code is invoked.