Amazon Robotics Challenge 2017 Datasets

Amazon Robotics Challenge 2017 Datasets includes two datasets: Image Matching Dataset and Grasping Dataset.

The Image Matching Dataset is a small and simple dataset with RGB-D images and height maps of various objects in a bin with manually annotated suckable regions and parallel jaw grasps, containing observed RGB-D images and product images of 61 different objects: 41 for training and testing, and 20 for testing only.

The Grasping Dataset is a set of RGB-D images containing grasped objects and representative product images from Amazon against a green screen, including two sub-datasets for suction and for parallel jaw grasping.

Both datasets were captured using an Intel® RealSense™ SR300 RGB-D camera. Color images were saved as 24-bit RGB PNGs; depth images and height maps were saved as 16-bit PNGs, with depth values saved with 10mm accuracy.

Invalid depth is set to 0 and the depth image is aligned with its corresponding color image.

Amazon Robotics Challenge 2017 Datasets was released by The MCube Lab, Princeton Vision and Robotics in 2018, and the main publisher was Andy Zeng.