Web server for image classification

Download the repository

Use git to clone the repository:

git clone https://github.com/unica-isde/web-server-2023

Optional but recommended - create conda environment: https://docs.conda.io/projects/miniconda/en/latest/

conda create --name isde python=3.10
conda activate isde

And install the requirements with

pip install -r requirements.txt

Additional requirements:

• Redis - install redis and check that it is running

Configuration

Configure the service by editing the file app/config.py.

Prepare the resources

It is recommended to pre-download images and models before running the server. This is to avoid unnecessary waits for users.

Run python app/prepare_images.py and python app/prepare_models.py. Models will be stored in your PyTorch cache directory, while the path for the image directory can be found in the app/config.py file.

python app/prepare_images.py
python app/prepare_models.py

Usage

Run locally

To run the code without containers, it is sufficient to run separately the server,

uvicorn main:app --reload

And then run the worker.py script. The worker will process jobs stored in the queue. In order for the queue to work, you should have redis
installed and running (specify port in config.py).

NOTE: on MacOS, it is necessary to disable the fork safety before running the worker. Don't worry, it will be disabled only for the current session of the terminal.

export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES