Deep Learning-Based Depth Estimation from Light Field

Light fields enable precise and robust depth estimation. With the shift from consumer light field cameras, new applications now require high-resolution, wide-baseline camera arrays and reliable confidence measures. This thesis introduces EPI-Shift, a deep-learning framework effective for both small- and wide-baseline light fields, and three deep learning methods for multimodal posterior regression. These methods extend the applicability to wide-baseline light fields and enhance posterior regression, performing on par with state-of-the-art approaches.