Rosetta: Understanding text in images and videos with machine learning

What this post added

This post introduces Rosetta, a large-scale machine learning system designed to extract and understand text from images and videos across a wide variety of languages. This capability directly enhances accessibility by enabling more relevant photo search and improving screen reader functionality for visually impaired users. It also contributes to community safety by helping to identify inappropriate content. The post details the technical approach, including text detection using Faster R-CNN and text recognition using a CNN with CTC loss, as well as the use of synthetic data generation for training.