Texture-less Object Detection Dataset (2013)

This dataset includes 9 texture-less models and 55 test scenes with clutter and occlusions. It was acquired with a webcam and comes with hand-labeled groundtruth for the pose of each model instance in the scene.


3D Keypoint Detection Dataset (2012)

This dataset, divided into 5 parts, can be used for evaluating 3D keypoint detectors’ performance. The whole dataset comes with groundtruth.


3D Object Recognition and Reconstruction Dataset (2011)

Here we provide a dataset for 3D object recognition, composed by a set of 3D models and a number of 3D scenes characterized by clutter and occlusion where the models have to be found.  Groundtruth for pose estimation is also available. The dataset is divided into 5 subsets:

  • two obtained from the models belonging to the Stanford Repository
  • two acquired by us using Spacetime Stereo
  • one acquired by us using the Kinect sensor

In addition, we provide a reconstruction dataset (both including Spacetime Stereo and Kinect data) including multiple views of the same objects and registration ground-truth.


Pattern Matching (2011)

This dataset includes several hundreds images and patterns with different sizes and nuisances (Gaussian noise, image blur, JPEG compression). The dataset has been proposed for the evaluation of pattern matching algorithms presented in:

“W. Ouyang, F. Tombari, S. Mattoccia, L. Di Stefano, W.K. Cham, “Performance Evaluation of Full Search Equivalent Pattern Matching Algorithms “, TPAMI (in print)


3D Semantic Segmentation (2011)

This dataset has been acquired with a Microsoft Kinect sensor and includes RGB-D data of 5 categories of common grocery products such as packets of biscuits, juice bottles, coffee cans and boxes of salt, of different brands and colors. The provided groundtruth allows evaluating semantic segmentation algorithms based on shape and/or RGB-D data.


3D Urban Dataset (2011)

This dataset has been obtained from the “3D Urban Challenge” data provided by prof. Ioannis Stamos. The original data has been divided into training, validation and test sets, based on 3 different categories. Inquiries about this dataset can be made via email.


Stereo Dataset (2010)

A stereo dataset composed of 6 stereo-pairs with groundtruth, acquired under realistic illumination conditions. It can be used to evaluate the performance of stereo matching algorithms.


Robust Pattern Matching Dataset (2008)

This dataset includes a set of templates and a set of images where the position of the templates has to be found. Images are affected by strong photometric distortions, noise (due to poor camera sensors) and small point of view variations. Groundtruth is also available. It can be used to evaluate the performance of pattern matching algorithms and robust matching measures.