COMPUTATIONAL VISION AT CALTECH

3D Objects on Turntable

[ Description ][ Download ][Other Datasets]

CamerasSetup
TopView1
TopView2

Description

This is a database of objects viewed from 144 calibrated viewpoints under 3 different lighting conditions. The objects are placed on an automated turntable and photographed every 5 degrees. Each object is photographed by 2 cameras arranged in a stereo setup, the pictures have a high 3 M-pixels resolution acquired by two Canon Powershot G1 digital cameras. This README file provides more details.

Download

This is a LARGE dataset (~60GB for the whole dataset), so you will probably want to download only part of it. This is why we did not zip it into a large archive. One option is to send us a physical hard drive, we will be happy to load it with the dataset and mail it back to you. The address is: Vision Lab attn:Pierre Moreels, Caltech MS136-93, 1200 East California Blvd, Pasadena CA91125.

If you would rather download the files over the network, we recommend the use of the command wget. For example, the command
wget -nH --cut-dirs=5 --mirror --no-parent http://www.vision.caltech.edu/pmoreels/Datasets/TurntableObjects/ImageSet070504/14.6mmZoom/Car/
will download the images of the Car object obtained by both cameras, and store them under your local directory in ./Car/Bottom/ and ./Car/Top

The directories containing the dataset are:
ImageSet070504/
ImageSet071204/
ImageSet071404/
ImageSet071604/
ImageSet072104/
ImageSet072204/
ImageSet072304/
ImageSet072604/
ImageSet072704/
ImageSet072904/
ImageSet080304/
ImageSet080404/
ImageSet091504/
As mentioned in the README file, the reason for having separate directories is that the Canon software controlling the cameras was crashing, forcing us to turn the cameras off and on, and to acquire a new set of calibration images.

Set of objects photographed with 4 different focal lengths:
4ScaleImageSet/
Set of objects photographed every degree (instead of every 5 degrees)
OneDegreeImageSet/
Movies showing most objects - these are just .avi files generated in matlab with the .jpg images in the above folders.
MoviesNorm/

Contributors

The dataset was collected in 2004 by Pierre Moreels and Javier Alatorre. We acknwledge the help of Dirk Neumann (Caltech) and Andreas Wenger (USC) for the code used to automate the system. this dataset:

Literature

Papers reporting experiments on this dataset:

1. Evaluation of Features Detectors and Descriptors based on 3D objects P. Moreels and P. Perona. ICCV 2005

2. Evaluation of Features Detectors and Descriptors based on 3D objects P. Moreels and P. Perona, IJCV 2006.

How to Reference this Dataset

We would appreciate it if you cite the above papers when using the dataset:


Last updated: July 3, 2006 by pierre at caltech dot edu.

Maintained by pierre at caltech dot edu