tx2_fcnn_node: An open-source ROS compatible tool for monocular depth reconstruction

摘要

We present tx2_fcnn_node – a Robot Operating System (ROS) compatible tool that is aimed at seamless integration of various monocular depth reconstruction neural networks to the robotic software based on ROS (which is a de-facto standard in the area of robotics). Our tool simplifies the process of deploying, evaluating, and comparing depth reconstruction neural networks both on real robots and in simulation. We complement our software with a set of the precompiled neural networks which can be used off the shelf, with some of them being able to demonstrate near real-time performance when running onboard compact embedded platforms, e.g. Nvidia Jetson TX2, that are often used nowadays both in academia and industry.