CLIP + FastAPI + Streamlit

Deployment of CLIP model FastAPI + Docker + Streamlit to recognise hotdog pictures.

Virtual Environment

Firt we need to create a virtual environment for the project, to keep track of every dependency, it is also useful to use and explicit version of Python

Install the package for creating a virtual environment: $ pip install virtualenv

Create a new virtual environment $ virtualenv venv

Activate virtual environment $ source venv/bin/activate

Python packages

Now with the virtual environment we can install the dependencies written in requirements.txt

$ pip install -r requirements.txt

Web application

Finally we can test our web application by running:

$ uvicorn main:app

Docker

Now that we have our web application running, we can use the Dockerfile to create an image for running our web application inside a container

$ docker build . -t hotdog_or_not

And now we can test our application using Docker

$ docker run -p 8000:8000 hotdog_or_not

Test!

Test by using the calls in tests/example_calls.txt from the terminal

or run in terminal $ cd tests $ python image_request.py

Run Streamlit app

$ streamlit run hotdog_app.py and upload a picture to get a prediction.

Evaluate clip model

Download images

To download images for testing there is a python script that allows us to search for a query and obtain images. It can be used from the terminal.

$ python get_test_images.py --query "{}" --max_images {}

Run evaluation

$ python clip_hotdog_evaluation.py