3DDL
Link to github project page: here
This is a project I worked on when I was pursuing a Masters in Computer Science at Imperial College London. This project was undertaken as an industrial project required to fulfill our degree.
The problem we had was to train a reliable classifier for grocery products for Ocado Technology, without much reliable data. Most of the images we had were from very generic views online, or some poorly taken images from a warehouse facility.
We managed to bring together a solution that takes in an actual instance of a grocery product, scan the product and generate a 3D model, and use a graphics engine to generate an unlimited amount of training images. We the trained an object classifier and an object detector, and developed a mobile app that we could use to help people re-order a grocery product from the Ocado website from an image of the product.
The technologies we used included photogrammetry (3DF Zephyr), graphics processing (Blender), and various deep learning libraries (Tensorflow, Keras).