Deep Dive

Ehsan Sayyad, Tobias Höllerer, Pradeep Sen

Environment
3rd floor

Deep Dive is a set of experiments in predicting depth from a single image using deep neural networks. A randomized dataset of textured 3D scenes and corresponding surface structures is created in different styles and is used to train multiple convolutional neural networks. This piece exhibits the dataset, learning process, and final results in a 3D real-time experience.