[Personal Robotics main page][Main Page]

Office Object Database

These are datasets collected and used for the paper
"A Fast Data Collection and Augmentation Procedure for Object Recognition",
Benjamin Sapp, Ashutosh Saxena, Andrew Y. Ng. AAAI 2008. [pdf]

The dataset contains images of 10 common office objects (cell phones, forks, hammers, keyboards, coffee mugs, pliers, scissors, staplers, telephones and watches) collected in 2 ways:

real-world data was collected by capturing the object in a natural office environment.
synthetic data was collected by capturing the object on a green screen background.

Browse the real-world data (real.tar.gz)
Browse the synthetic data (synthetic.ta.gz)

Both datasets contain all 10 object classes, with 200 images per class. The real data includes images and corresponding text files which give bounding boxes for the object. The synthetic data includes images, bounding box information, foreground masks, and shadow masks that result from our methods, titled appropriately. In all cases, the format of the text files specifying bounding boxes are one line, with 4 numbers delimited by spaces. The four numbers correspond to the top left and bottom right (x,y) locations of the bounding box: [left top right bottom], i.e., [pt1.x pt1.y pt2.x pt2.y].