Deep Learning for Depth Learning CS 229 Course Project, 2013 Fall

Extracting 3D depth information from images is a classic problem of computer vision. Traditionally image depth could be extracted by techniques such as stereo camera or images from multiple views. In this project, we are trying to recognize the depth information by using a single still image from single camera, which has great potential applications in vision and recognition. To learn the complex relationship between single RGB image with its depth information, we chose to use Deep learning algorithms, to learn the multiple level features and corresponding different levels of abstraction. The main goal of the project is to train a deep network that is able to extract local and non-local features and predict a depth-map given a still image. In addition, implementation of other learning algorithm (kernel based, dictionary learning) was also conducted for comparison.