UC San Diego

This thesis represents a novel approach to augment data for unsupervised depth estimation. Data augmentation is an effective method to improve the performance of neural network models. However, such data augmentation strategies need to be hand-designed for the task for which the network is trained. In this work, we focus on learning data augmentation strategies from data itself using population-based training. We show the effectiveness of this approach in unsupervised learning setting for depth estimation task in monocular videos on KITTI dataset.