README.md

MLFlow_fasterrcnn

Getting started

Object detection using FasterRCNN model Introduce MLFlow as a solution:

• to track all experiments during model training (see mlflow_train.py code)
• store the best trained model in the Model Registry (see mlflow_train.py code)
• serve the model (after deployment) to be consumed by users for prediction purposes on new data. (see mlflow_inference.py code)

To launch it, run the following commands:

git clone https://github.com/falibabaei//fasterrcnn_pytorch_api
cd fasterrcnn_pytorch_api
pip install -e .
git submodule init
git submodule update
cd submodule/dir
pip install -e .
cd ..

The associated Docker container for this module can be found in https://github.com/falibabaei//DEEP-OC-fasterrcnn_pytorch_api.

You might need to install some other dependent packages/libraries as follows:

pip install setuptools wheel opencv-python gcc libgl1  vision_transformers albumentations

Prerequisites

Main installation you have:

• python3 > 3.7 but lower than 3.11
• pip is installed

Install mlflow framework

pip install mlflow[extras]

Configuring MLFlow constants

Remote MLFlow server MLFLOW_REMOTE_SERVER="http://mlflow.dev.ai4eosc.eu"
#Set the MLflow server and backend and artifact stores
mlflow.set_tracking_uri(MLFLOW_REMOTE_SERVER)

for direct API calls via HTTP we need to inject credentials
MLFLOW_TRACKING_USERNAME = 'mlflow_user'
MLFLOW_TRACKING_PASSWORD = getpass.getpass() # inject password by typing manually

for MLFLow-way we have to set the following environment variables
os.environ['MLFLOW_TRACKING_USERNAME'] = MLFLOW_TRACKING_USERNAME
os.environ['MLFLOW_TRACKING_PASSWORD'] = MLFLOW_TRACKING_PASSWORD

Acknowledgment

This work is co-funded by AI4EOSC project that has received funding from the European Union's Horizon Europe 2022 research and innovation programme under agreement No 101058593

License

For open source projects, say how it is licensed.