Open BenjaminYoung29 opened 5 years ago
Thank you for your interest! The code is so far only tested in simulation with a virtual robot. However it is implemented in ROS + Tensorflow so it could theoretically work on a real robot that has those frameworks installed. It will require some adjustments depending on the type of laser scanner and scanning motion.
Thanks for your comment. I am a newer to 3d point net. I wonder what point net segmentation can do in real life at this point that 2d segmentation can not do. Because it seems that all the paper I've read are comparing accuracy and my tutor won't allow me doing that:(
3D Point Nets are relatively newer in development, but has a lot of important applications in autonomous driving, 3D mapping, and any robot that uses laser scanners. The main advantage of 3D segmentation compared to 2D segmentation is that the real physical size and shape of detected objects can be measured accurately
I've read your paper and find that you run it on virtual mobile robot with ROS? So GPU is still usable. However is the framework still work on RPI?