abhishekrs4 / overhead_mnist

DL model deployment using docker, API deployment with FastAPI, and MLOps using WandB for overhead-mnist dataset

Overhead MNIST

Dataset info

• The dataset used in this project is the Overhead-MNIST dataset and it can be found here
• The dataset contains 10 classes

Repo

• This repo contains an end-to-end deep learning project deployment for overhead image classification
• Weights and Biases has been used for the MLOps
• For deployment, an API has been developed and deployed using FastAPI and docker
• For the training, the dataset is split into 95% - 5% for train and validation sets respectively
• The python packages are listed in requirements.txt
• The docker container can be deployed using Dockerfile
• For training and logging the model, use the modeling/train.py script
• The FastAPI app deployment code is in app.py script
• To test the deployed FastAPI app on a local machine, the test_post_request.py script can be used
• Some sample test images are available in sample_test_images

Docker deployment instructions on a local machine

• For deployment requirements_deployment.txt needs to be used
• To build the container, run the following command

docker build -t fastapi_overhead_mnist .

• To the run the container, run the following command

docker run -p 7860:7860 -t fastapi_overhead_mnist

HuggingFace deployment

• The FastAPI application has also been deployed to HuggingFace
• To test the deployed FastAPI app on HuggingFace, use the test_post_request.py script in the HuggingFace repo since the endpoint is different

Docs

• The docs generated with sphinx can be found in _build/html/index.html

Sample test images