eigengravy/lensq

Image Processing API built w/ Flask, Keras and Celery.

lensq

lensq is a application that provides an API for predicting handwritten digits using a pre-trained deep learning model.

Built w/ Flask, Keras and Celery.

Setup Development Environment

1. Clone the repository:

git clone https://github.com/your-username/lensq.git
cd lensq

2. Install dependencies and drop into env shell:

poetry install
poetry shell

3. Download the pre-trained model weights file model.keras and place it in the project root directory. Alternatively, you can train it locally using scripts/train.py.

Model

The application uses a pre-trained CNN model for handwritten digit recognition. The model is trained on the MNIST dataset and achieves an accuracy of around 99% on the test set.

The pre-trained weights model.keras are required to run the application. You can download the weights file from the project repository or train the model yourself using the provided train.py script.

Running the Application Locally

1. Spin up a redis instance using docker run on port 6379. Alternatively, you can run redis locally.

docker run -d -p 6379:6379 redis 

2. Start the celery worker.

celery -A make_celery worker --loglevel INFO  

3. In a separate terminal, start the Flask application:

flask -A lensq run --debug

The application should now be accessible at http://localhost:5000.

Deploying to Server

To deploy the application to a server environment, follow these steps:

1. Setup an EC2 instance and install required dependencies (Redis, Celery, Flask) on the server.
2. Clone the repository and copy the project files to the server.
3. Place the pre-trained model weights file model.keras in the project root directory.
4. Configure the Redis and Celery settings in lensq/__init__.py as per your server environment.
5. Start the Redis server.
6. Start the Celery worker.
7. Start the Flask application using Gunicorn.

pm2 is used to manage the celery and flask jobs.

API Documentation

Endpoint: /tasks/predict

Method: POST

Input Parameters:

• images (multipart/form-data): One or more image files to be processed for digit recognition.

Response:

{
    "result_id": "string"
}

The result_id is a unique identifier for the submitted task. It can be used to check the status and retrieve the prediction results.

Endpoint: /tasks/result/<id>

Method: GET

Path Parameter:

• id (string): The result_id returned by the /tasks/predict endpoint.

Response:

{
    "ready": bool,
    "successful": bool,
    "value": list
}

• ready: Indicates whether the prediction task has completed.
• successful: Indicates whether the prediction task was successful (only present when ready is true).
• value: A list of predicted digit labels for each submitted image (only present when ready and successful are true).