Yolo Training Pipelines

This repo is in progress, it explores different options to train yolo for a data set of Uno Cards images.

At some point it should cover the whole lifecycle of training a yolo model:

• Experimentation using Jupyter
• Externalize code from Jupyter to python files and functions
• Create a training Kubeflow pipeline, including input and output of the models
• Use MLFlow to track metrics and models
• Use Optuna for fine tuning hyperparameters
• Simplify with container components
• Serve the model with KServe/KFServing

So far most of the work is than but scattered...

Buckets needed

datasets mlflow models pipelines

Upload datasets

Upload these files to s3://minio//datasets

• uno-cards-v1.1.zip
• uno-cards-v1.2.zip

Upload yolo model

Upload yolo to s3://minio//ultralytics

Refererences

https://learnopencv.com/ultralytics-yolov8/ https://learnopencv.com/train-yolov8-on-custom-dataset/ https://docs.ultralytics.com/integrations/mlflow/#how-to-use

https://rh-aiservices-bu.github.io/parasol-insurance/modules/04-05-model-serving.html