ARC2017 RGB-D Dataset for Object Detection and Segmentation
- Author
- R. Araki, T. Yamashita and H. Fujiyoshi
- Publication
- Late Breaking Results Poster on International Conference on Robotics and Automation, 2018
Download: PDF (English)
In July 2017, Amazon Robotics LLC held a competition called the Amazon Robotics Challenge (ARC) in Nagoya City, Japan, to compare different logistics automation technologies. Contestants had to complete two tasks — a Stow task, where robots were required to stow 40 different types of items from a tote into a storage system, and a Pick task, where they had to pick items from the storage system and place them in the cardboard boxes. In the final round, these two tasks had to be performed consecutively. The items included various rigid, non-rigid, and translucent objects.
We competed as part of the Team MC^2 and created our own dataset for the ARC2017 tasks. This dataset comprises images of the 40 items used in the ARC 2017 competition that were in a red tote. For every scene, the dataset includes an RGB image, a depth map image, and correctly labeled bounding-box and segmentation data.