Updates for cmltrain (deep learning with python via cmltrain), added facecrop plugin

[tobiasbaur]

• still needs proper way to capture video resolution via meta when applied in live (supports only standard resolutions for now)