bart: blocking ads in real time

This project aims to simply detect ad billboards in the user's camera. Think of it as the starting point for a real life ad blocker.

BART consists of two components: training models for billboard detection, and deploying them live for use on the web at https://bart.fly.dev/

The website (src) serves as a simple way of testing the different models trained on the yolo8 directory. It lets you test the different models inside the src/model dir and allows you to select the image transformation to apply on the detections (such as blurring the video wherever a billboard appears). It also provides some profiling tools to test the accuracy and performance of each model.

The model training directory (yolo8) contains several tools to train the experiments with different parameters and get the final models for the website to use.