Requested feature to use a `custom_model` per camera?

[wjbridge]

First off, let me say that this opensouce project is amazing! This has streamlined my AI processing integration with HA a ton! I was experimenting with the Codeproject AI block. I have one camera that I want to detect a cat object and another camera that I want to detect a postal package so I can get a notification.

Using the default dataset coco, I can detect the cat as this model has been trained to do this. I separately trained a custom model to detect postal packages but I realized that you can't configure Viseron to use this model for a particular camera. I know I can train a custom model with all objects but I don't have the hw resources to do this easily. This led me to the question if it would be possible to map the custom_model to a camera for the codeprojectai component?

Proposed Config:

codeprojectai:
  host: !secret host
  port: !secret port
  object_detector:
    cameras:
      basement_cat:  # Attach detector to the configured camera_1 above
        fps: 1
        custom_model: ip_general
        scan_on_motion_only: true  # Scan for objects even when there is no motion
        zones:
          - name: cat_litter_box
            coordinates:
              - x: 134
                y: 504
              - x: 794
                y: 488
              - x: 794
                y: 1080
              - x: 134
                y: 1080
            labels:
              - label: cat
                confidence: 0.75
                trigger_recorder: true
      front_door_package:
        fps: 1
        custom_model: parcel
        scan_on_motion_only: true  # Scan for objects even when there is no motion
        labels:
          - label: parcel
            confidence: 0.7
            trigger_recorder: true

[roflcoopter]

Great to hear! Yes I think that should be possible, will investigate further

[aliyoge]

Any progress?

[roflcoopter]

Sadly not had the time to look at this just yet!