This project was all about developing a model that could effectively detect and obscure car license plates in images. This stemmed from a prior endeavor that successfully removed backgrounds from car images. The goal was to ensure privacy by rendering license plates unrecognizable in pictures of vehicles.
I utilized Keras, a popular open-source neural-network library, to construct and train the model. Keras, with its user-friendly interface and flexibility, was instrumental in the model's development process.
Downsizing and Deployment
One of the crucial tasks in this project was to streamline the code and the model's weight to enable efficient deployment in serverless environments. After significant work in optimization and reduction, I managed to deploy the model using Google Cloud Functions, a part of Google Cloud Platform (GCP).
Serverless deployment like this has some significant advantages. For one, it takes care of scalability, which is a big relief as it automatically adjusts resources based on demand. Moreover, it's cost-effective. In my case, I got a pretty good deal, with up to one million operations per month free of charge.
So, that's the gist of it: a compact, serverless model capable of detecting and obscuring car license plates, all thanks to Keras and Google Cloud Functions.
Feel free to try out the model which i've deployed on my virtual lab of sorts called ITERATIVE Code.
You can find it here: https://licenseplate-ai.iterativecode.io/