Implements a web app to which images can be uploaded. A TensorFlow model is then used to perform object detection. The model is served through TensorFlow Serving. The web app and the TF-Serving server is packaged using Docker. Docker-compose is used to create a virtual network for both container such that the Flask app can access the TF Serving container (in this case via REST).
- root
- .env
- Defines several environment variables that are used within the docker-compose file and the web app
- docker-compose.yml
- Defines the services (Flask web app + TF Serving) such that they are run together. Utilises the GPU (prerequisites).
- docker-compose_CPU.yml
- Same as the file above but doesn't utilise the GPU (and, therefore, also doesn't require any extra prerequisites).
- .env
- flask_app
- Contains the Flask web app along with the respective Dockerfile.
- models
- Contains a TensorFlow model from the TensorFlow object detection model zoo. EfficientDet D4 is used for this project.
To run the entire app, type (in the same directory as this README file):
sudo docker-compose up
Wait a few seconds until both containers are spun up. Then visit localhost:8500 to view the app.
To use the app select a valid image (.jpg, .jpeg, .png) and hit "Submit". The result will be displayed within a few seconds.
