Using Compute vision in dental radiograph

DANIAH MANSOUR ATTIAH ALHAZMI

Backend models from Azure Custom vision are way different from Yolo v8. We can export the Azure Custom vision models as Tensorflow/pytorch models but I don't think it will be a good practice a train a model in Custom vision framework and transfer to Yolo v8 model.

Best practice would be fine tune yolo v8 model in Azure ML studio (GPU environment) with a custom dataset.

https://github.com/airacingtech/YOLOv8-Fine-Tune

https://yolov8.org/how-to-use-fine-tune-yolov8/

Fine tuning scripts can be run in Azure ML GPU or Local GPU, Post finetuning, we can upload the model registry in Azure ML

https://learn.microsoft.com/en-us/azure/machine-learning/tutorial-deploy-model?view=azureml-api-2

You can go through authoring scoring script to deploy the fine tune model as online or batch endpoint on Azure ML environment instead

https://learn.microsoft.com/en-us/azure/machine-learning/how-to-batch-scoring-script?view=azureml-api-2&tabs=cli

Scoring happens on endpoint in two steps.

1. Load the fine tune models from model registry

      def init():
          global model
          # AZUREML_MODEL_DIR is an environment variable created during deployment
          # The path "model" is the name of the registered model's folder
          model_path = os.path.join(os.environ["AZUREML_MODEL_DIR"], "model")
          # load the model
          model = load_model(model_path)
   

2. Load the dataset and run prediction with endpoint in batches

import pandas as pd
from typing import List, Any, Union
def run(mini_batch: List[str]) -> Union[List[Any], pd.DataFrame]:
    results = []
    for file in mini_batch:
        (...)
    return pd.DataFrame(results)

Hope it helps.

Thank you