Abstract: Grasp detection research focuses at the moment on finding neural networks that given a RGB-D image or point cloud, yield a parametric grasp description that can be used to firmly grip target objects. There is a need for these models to be small and ecient, such that they can be used in embedded hardware. Furthermore these models tend to only work for top-down views, which highly restrict the ways objects can be grasped. In this work, we focus on improving an existing shallow network, GG-CNN, and propose a new orthographic pipeline to enable the use of these models independently of the orientation of the camera. We make our implementation available on GitHub.