3D Object Recognition Dataset

Akash Kushal, Jean Ponce

Department of Computer Science, University of Illinois, Urbana Champaign

This page describes a 3D object recognition dataset. The dataset consists of 9 objects and 80 test images. The training images are stereo views for each of the 9 objects that are roughly equally spaced around the equatorial ring for each of them. The number of stereo views ranges from 7 to 12 for the different objects. The test images are monocular images of objects under varying amounts of clutter and occlusion and different lighting conditions.

Objects Used:

Salt 8 stereo pairs
Vase 8 stereo pairs
Yogurt 8 stereo pairs
Bournvita 8 stereo pairs
Bear 8 stereo pairs
Ball 12 stereo pairs
Chest-Buster 7 stereo pairs
Green-Dragon 12 stereo pairs
Small-Dragon 12 stereo pairs

Browse or Download the Dataset

Archive in .tgz format  (~111 MB)

Image Directory

Note: Each object directory contains a file named camerafile.txt which contains the left and right 4x3 projection matrices for the two stereo pairs in that directory. Each test image's name has as substrings the names of all the objects it contains. The training images are color JPEGs with resolution 8 Mpix (3504x2336) and the test images have resolution 2 MPix (1728x1152).

Our Results

Our results on this dataset are presented in the following paper

Modeling 3D objects from stereo views and recognizing them in photographs. (PDF)
Akash Kushal, Jean Ponce Proc. European Conference on Computer Vision, 2006.

The creation of this dataset was supported in part by the National Science Foundation under grants IIS 03-12438, IIS 03-08087 and IIS 05-35152, the UIUC-Toyota collaboration for 3D object modeling, recognition and classification from photographs, and the Beckman Institute.