Installation serverless YOLOv8 functions on Windows 10 with using the Ubuntu subsystem

• Navigate to the directory "cvat/serverless/yolov8/nuclio". (If it does not exist, create it.) Copy the necessary files to the directory with the following command.

git clone https://github.com/zahidesatmutlu/custom-yolov8-model-cvat-auto-annotation
cd custom-yolov8-model-cvat-auto-annotation

• Update the "metadata/annotations/spec" section in the "function.yaml" file with the classes you trained the model with. For example:

metadata:
  name: custom-model-yolov8
  namespace: cvat
  annotations:
    name: custom-model-yolov8
    type: detector
    framework: pytorch
    # change this accordingly to your model output/classes
    spec: |
      [
        { "id": 0, "name": "car" },
        { "id": 1, "name": "person" },
        { "id": 2, "name": "motorcycle" },
        { "id": 3, "name": "truck" }
      ]

• Copy your model weight in .pt format to the same directory ("cvat/serverless/yolov8/nuclio"). In "main.py" insert the model weight you trained in "model = YOLO('best.pt')" (Line 13).

• Go back to the "/cvat" directory and deploy the YOLOv8 functions.

./serverless/deploy_cpu.sh serverless/yolov8/nuclio

or

./serverless/deploy_gpu.sh serverless/yolov8/nuclio

• Open Google Chrome browser and navigate to localhost:8080. You can now view the custom YOLOv8 model for CVAT!

YOLOv8 model for CVAT