The berkeley 3d object dataset

The task of object recognition has made significant advances in the past decade and crucial to this success has been the creation of large datasets. Unfortunately, these successes have been limited to the use of intensity images and have chosen to ignore the very important cue of depth. Depth has long been thought to be an essential part of successful object recognition, but the reliance on large datasets has minimized the importance of depth. Collection of large datasets of intensity images is no longer difficult with the wide spread availability of images on the web and the relative ease of annotating datasets using Amazon Mechanical Turk. Recently, there has been a resurgence of interest in available 3-D sensing techniques due to advances in active depth sensing, including techniques based on LIDAR, time-of-flight (Canesta), and projected texture stereo (PR2). The Primesense sensor used on the Microsoft Kinect [4] gaming interface offers a particularly attractive set of capabilities, and is quite likely the most common depth sensor available worldwide due to its rapid market acceptance (8 million Kinects were sold in just the first two months). There is a large body of literature on instance recognition using 3-D scans from the computer vision and robotics communities. However, there are surprisingly few existing datasets for category-level 3-D recognition, or for recognition in cluttered indoor scenes, despite the obvious