Print
Univis
Search
 
FAU-Logo
Techn. Fakultät Website deprecated and outdated. Click here for the new site. FAU-Logo

Data

These are datasets that our group used in our publications and which we offer free of charge for academic purposes (we provide free software, too).

Computational Two-Illuminant Datasets

Obtaining quantitative ground truth for color constancy under multiple light sources is difficult. We provide two datasets of 58 and 20 images that are exposed to two illuminants. The ground truth is obtained from an algorithm that operates on multiple aligned input images. Opens internal link in current windowRead more & download...

Carotid Artery TOF MRA Data Set (PASCAL)

The segmentation of the carotid bifurcation region in non contrast-enhanced MRA TOF datasets is challenging due to its susceptibility to irregular blood flow patterns. These lead to intensity deviations especially close to the bifurcation and even to signal voids.

We provide here standardized MRA TOF datasets of this region along with manual segmentations to be used to evaluate segmentation methods.

The datasets can be found Opens internal link in current windowhere.

DaLiAc - Daily Life Activities Datasets

Various approaches for movement analysis have been proposed in literature. Up to date, it is not clear which solution is outperforming the others. The provision of benchmark datasets are mandatory for research groups. This could enhance and speed up the process of getting the best performing and most appropriate algorithm for movement analysis. Currently, in our database are examples for daily life activity recognition and energy expenditure estimation. 

The datasets can be found Opens internal link in current windowhere.

 

 

Digital Brain Perfusion Phantom

The Digital Brain Perfusion Phantom package provides data and Matlab tools to create a realistic digital 4D brain phantom, in particular for reproducible evaluation of reconstruction algorithms using non-linear regularization for perfusion CT and perfusion C-arm CT. It is based on the Realistic Digital Brain Perfusion Phantom originally published by Riordan et al. [1]. Since classical digital CT phantoms usually consist of homogeneous structures and have very sparse representation in transformations like total variation or wavelets, they highly favor non-linear reconstruction algorithms. This phantom allows for a more authentic evaluation by providing a brain model based on real physiological data and avoiding sparsity by continuously varying perfusion parameters and anatomical structures using MR data.


The phantom is available for Opens internal link in current windowdownload here.

Multi-Illuminant Dataset

Color constancy algorithms under non-uniform illumination are particularly hard to evaluate. Ground truth illumination measurements are either inaccurate or only sparsely set in the scene. We provide a dataset of four scenes, captured under combinations of 6 spectral filters. After the capturing, the scenes were painted gray and recaptured. This protocol provides pixel-wise ground truth, but discards interreflections. We assumed this to be a reasonable trade-off to evaluate our algorithms. Read more & download...

Gold Standard Database for Evaluation of Fundus Image Segmentation Algorithms

This database has been established by a collaborative research group to support comparative studies on automatic segmentation algorithms on retinal fundus images. The database will be iteratively extended and the webpage will be improved. We would like to help researchers in the evaluation of segmentation algorithms. We encourage anyone working with segmentation algorithms who found our database useful to send us their evaluation results with a reference to a paper where it is described. This way we can extend our database of algorithms with the given results to keep it always up-to-date.
The database is available for download here.

Supervised Multispectral Image Segmentation Dataset

In this dataset we provide 32 segmentation tasks in nine images of the CAVE Multispectral Image Database. The dataset includes hand-labeled ground-truth data and seed-points for each task as well as segmentation results of several algorithms.

The dataset can be found Opens internal link in current windowhere.

Image Manipulation Dataset

A ground truth dataset for benchmarking the detection of image tampering artifacts.

This dataset contains 96 images in various stages from original to manipulated, as well as the respective tampered parts. Also available is a software framework for a fine-grained control over combinations of various manipulations. Thus, it is for instance possible to create differently JPEG-compressed, visually credible forgeries suitable for evaluation of a wide range of image forensics algorithms.

The dataset is available for download here.

CAVAREV: CArdiac VAsculature Reconstruction EValuation

The main goal of CAVAREV is to enable an easy and objective comparison of different dynamic reconstruction algorithms.

The area of application is the 3-D and 4-D reconstruction of contrasted cardiac vessels, e.g. the coronary arteries using C-arm CT (rotational angiography). Various methods exist in literature with lots of nice results. However, the results can vary significantly between phantom and real clinical data.

Therefore, we provide:

  • Open (i.e. public) and easy accessible projection data.
  • Anatomically and physiologically correct projection data based on patient data.
  • Two degrees of difficulty: (a) Strictly periodic cardiac motion and (b) an aperiodic combination of cardiac and breathing motion.
  • An online evaluation platform where 3-D reconstructions can be uploaded and the reconstruction quality is assessed objectively.

For more information please visit www.cavarev.com.

RabbitCT: Benchmark for High-speed C-arm CT

RabbitCT is an open-source benchmark for C-arm CT FBP reconstruction.

It features a clincal C-arm CT dataset with 496 projections, and a software framework for measuring execution time and calculating reconstruction errors.
For more information please visit www.rabbitct.com.

An Annotated Image Database for Evaluation of Cell Detection Algorithms

We have created a database of five cell lines in order to assess the performance of cell detection algorithms. The database is composed of more than 3500 cells in 16 images of three real cell lines and 30000 cells in 200 images of two simulated cell lines.

Opens internal link in current windowDownload and more information

3-D Satellite Camera Frames

Here, we present the 25 frames used for situs reconstruction in our MICCAI publication ("3-D Operation Situs Reconstruction With Time-of-Flight Satellite Cameras Using Photogeometric Data Fusion"). The single frames are included as vtk files and the complete sequence with all frames is included as an ris file. Use Opens internal link in current windowRITK to open the ris file and visualize invalid pixels and the 3-D representation.

Initiates file downloadData archive

The following configuration file for ritk is used to visualize the sequence. Make sure to have all plugins available.

Initiates file downloadConfiguration file

 

 

Multi-Sensor Super-Resolution Datasets

Here, we provide the data sets used in our research on multi-sensor super-resolution for hybrid range imaging. In the current state, the database contains synthetic as well as real datasets acquired for different applications and modalities (e.g. hybrid 3-D endoscopy based on Time-of-Flight Imaging, 3-D data acquisition for indoor scenes using Microsoft Kinect).  More data sets to cover new applications in medical imaging and computer vision will be included in the future.

 

The whole database with a brief description of the image data is available here

Bamboo Scroll Dataset

Here, we provide two 3-D X-ray CT volumes and the segmented 32 slips of a wooden scroll bought in Beijing. The scroll is made of bamboo with carved writings and drawings. The first scan we made was a initial scan of the scroll without any soiling. As the scrolls are normally found underneath the earth in a poor condition, the scroll was heavily contaminated by potting soil, put in a plastic bag and scanned a second time.

The whole dataset with a detailled description of the volume data is available here.

DeepAL fieldwork data 2017/2018 (DLFD)

The DeepAL fieldwork data 2017/2018 (DLFD) was collected via a 15-m research trimaran in Northern British
Columbia (Vancouver Island) by an interdisciplinary team consisting of marine biologists, computer scientists and psychologists, adhering to the requirements by Department of Fisheries and Oceans in Canada. The entire dataset (raw data, labeled data) will be publicly available upon request only at the end of October 2019. For questions please contact Christian Bergler (christian.bergler@fau.de).