RGB-D Object Dataset
The RGB-D Object Dataset is a large dataset of 300 common household objects. The objects are organized into 51 categories arranged using WordNet hypernym-hyponym relationships (similar to ImageNet). This dataset was recorded using a Kinect style 3D camera that records synchronized and aligned 640x480 RGB and depth images at 30 Hz. Each object was placed on a turntable and video sequences were captured for one whole rotation. For each object, there are 3 video sequences, each recorded with the camera mounted at a different height so that the object is viewed from different angles with the horizon.
Unlike many existing datasets,such as Caltech 101 and ImageNet, objects in this dataset are organized into both categories and instances. In these datasets, the class dog contains images from many different dogs and there is no way to tell whether two images contain the same dog, while in the RGB-D Object Dataset the category soda can is divided into physically unique instances like Pepsi Can and Mountain Dew Can. The dataset also provides ground truth pose information for all 300 objects.
RGB-D Scenes Dataset v.2
- The RGB-D Scenes Dataset v2 consists of 14 scenes containing furniture (chair, coffee table, sofa, table) and a subset of the objects in the RGB-D Object Dataset (bowls, caps, cereal boxes, coffee mugs, and soda cans). Each scene is a point cloud created by aligning a set of video frames using Patch Volumes Mapping*. These 3D reconstructions and ground truth object annotations are exactly those used in our ICRA 2014 paper (see README).
RGB-D Scenes Dataset
- This dataset contains 8 scenes annotated with objects that belong to the RGB-D Object Dataset. Each scene is a single video sequence consisting of multiple RGB-D frames.
This part of the dataset contains the cropped RGB-D frames that tightly include the object as it is spun around on a turntable. There is another part of the dataset available containing 3D point clouds, in PCD format readable with the ROS Point Cloud Library (PCL), as well as a part containing the full 640x480 images from sensor.
This part of the dataset contains the 3D point clouds of views of each object, in PCD format readable with the ROS Point Cloud Library (PCL). There is a part of the dataset available containing cropped images of the objects for extracting visual features, and another part of the dataset containing the full 640x480 images from the sensor.
This part of the dataset contains the full 640x480 RGB-D frames. There is another part of the dataset available containing the cropped images of just the object on the turntable. There is also a part of the dataset with the 3D point clouds of views of each object in PCD format, readable with ROS Point Cloud Library (PCL).
This part of the dataset contains the ground truth pose labels for every image in the RGB-D Object Dataset, exactly as used in the pose recognition evaluation of the AAAI-11 paper (see README).
This part of the dataset contains the cropped RGB-D frames that tightly include the object, exactly as used in the object recognition evaluation of the paper introducing the RGB-D Object Dataset (i.e. subsampled every 5th video frame). Use this dataset if you wish to compare directly against object recognition results published in our papers. There is a separate download for the full dataset containing all RGB-D frames.
rgbd-scenes-v2_pc.zip Aligned scene point clouds, ground truth annotations, and camera pose estimates from 3D scene reconstruction
All RGB and depth image frames
objects_3dwarehouse.zip Point clouds of Trimble 3D Warehouse objects used for learning HMP3D features and classifiers in our ICRA 2014 paper, in PLY format (see README).