ML: Listen folder, detect & blur

[punov]

Create service to listen for folder /mnt/data/unprocessed_framekm

And once there's new FrameKM package,

• read every frame from it with the cursor
• run resize, inference & masking for every frame,
• prepare JSON file containing the following information for every processed frame:
  • resizing time
  • inference time
  • masking time
  • hash of model weights
  • class id
  • confidence
  • bounding box info (cx, cy, width, height)
  • with name of the JSON similar to name of the FrameKM but with '.json' extension
• packing all the processed frames into new frameKM in /mnt/data/framekm folder
• dropping metadata JSON into /mnt/data/ml_metadata

[Eduard-Voiculescu]

Currently there is not much error handling.

• add error handling when creating and moving files

Also, the python chime only works on newly created framekm files. Do we want to extend that to already existing framekms?