Machine Learning at Dropbox


Published on April 12, 2017

Machine learning powers many parts of our business at Dropbox, and is instrumental to the developent of our document scanner. The doc scanner makes it fast and easy for users get physical documents into their Dropbox. Using the Dropbox app on Android and iOS, users can convert a photo of their document into a clean, rectangular PDF or image. Our state-of-the-art Optical Character Recognition (OCR) system also lets Dropbox Business users search these scans by the text they contain. In a series of tech blog posts, we show how we used deep learning to build the document scanner and implement the OCR pipeline, and we share what we learned along the way.