Supervisor

Suitable for

Abstract

Diffusion-based generative models (such as stable diffusion) have shown that it is possible to generate photorealistic images from text input. For many VR/AR and robotics applications it is desirable to generate multiple images of the same subject or scene from different viewpoints. In this project you will investigate how to get diffusion models to generate multiple views of the same subject from different viewpoints. The consistency of the views will be evaluated by running them through a structure from motion system to obtain a full 3D model of the scene.

Prerequisites: Suitable for those who have done the Machine Learning course. Having also done Computer Graphics would be useful.