Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 27, 2020 - May 30, 2020
This paper presents a convenient, semi-automated pipeline for the annotation of real-world RGB-D data with pose ground truths of known objects using minimal input from the human user. This method enables rapid generation of labeled datasets for the training of CNN-based pose estimation of the objects desired in the user’s particular application.