Learning Pixel-Perfect 3D Vision and Generative Modeling

Generating images with the aid of computers has come a long way. Today’s technology and algorithms can simulate the world around us. What is more, the computer vision technique can recognise and predict identities and actions from pictures or videos. However, computer vision cannot manage 3D shapes correctly, and its semantics are not matched with pixel-perfect appearances. As a result, the designing of 3D environments, such as in games or films, remains laborious. The EU-funded PIPE project will work to solve these problems with new models that combine computer vision and simulation with machine learning for pixel-perfect 3D vision and generative modelling. With the use of deep convolutional neural networks learning, it will allow the creation of realistic samples of meaningful synthetic images.