In the past few years deep learning has emerged as a common approach to learning data-driven representations.  While deep learning approaches have attained remarkable performance improvements in many 2D vision tasks, such as image classification and object detection, they cannot be directly applied to geometric vision problems due to the fundamental differences between 2D and 3D vision tasks, such as the non-Euclidean nature of geometric objects, higher dimensionality, and the lack of large-scale annotated 3D datasets.  Designing geometric components and constraints to improve the performance of deep neural networks is a promising direction worth further exploration. The workshop aims to bring together experts from both the geometric vision and deep learning areas to summarize recent advances, exchange ideas, and inspire new directions.