Institut für Kognitionswissenschaft

Institute of Cognitive Science

Navigation und Suche der Universität Osnabrück



Osnabrück — Synthetic Scalable Cube Dataset; Voxel Based Dataset for Systematic 3D reconstruction by ANNs


Retrieving the 3D shape of an object from a collection of images or a video is currently realized with multiple view geometry algorithms, most commonly Structure from Motion (SfM) methods. With the aim of introducing artificial neuronal networks (ANN) into the domain of image-based 3D reconstruction of unknown object categories, we developed a scalable voxel-based dataset in which one can choose different training and testing subsets. We show that image-based 3D shape reconstruction by ANNs is possible, and we evaluate the aspect of scalability by examining the correlation between the complexity of the reconstructed object and the required amount of training samples. Along with our dataset, we are introducing, in this paper, a first baseline achieved by an only five-layer ANN. For capturing life’s complexity, the ANNs trained on our dataset can be used a as pre-trained starting point and adapted for further investigation. Finally, we conclude with a discussion of open issues and further work empowering 3D reconstruction on real world images or video sequences by a CAD-model based ANN training data set.


  Cube Setup Link
voxelcubes voxelcubes voxelcubes voxelcubes ... voxelcubes 3x3x3
100 000 cubes
Download - jpg views + txt
Download - obj files
voxelcubes voxelcubes voxelcubes voxelcubes ... voxelcubes 4x4x4
300 000 cubes
Download - jpg views + txt
Download - obj files
voxelcubes voxelcubes voxelcubes voxelcubes ... voxelcubes 8x8x8
430 000 cubes
Download - jpg views + txt
Download - obj files
This work is licensed under a Creative Commons Attribution 4.0 International License.
For giving appropriate credit, cite one of the publication from the references below.
Creative Commons License

Generator Tools

Python Voxelizer

This pyhton script create voxelized objects incl. a voxel set list out of ply, off or stl 3D object files.

Download Python Voxelizer

Cube Generator

This generator, written in Matlab, randomly generates n 3D objects. Each such object is created by taking a unit cube in R^3 and subdividing it into a r x r x r subcubes. The parameter r can be defined by the user. By ensuring the uniqueness of the cube distribution in the voxel grid, this generator is able to generate 2^(r^3) different 3D objects and export them as 3D *.obj object files.

Download Cube Generator

Views Generator

This generator, written in Matlab, (can be optionally used for voxelization of 3D objects and) renders w input images with a pixel resolution x by x. Where the w different viewpoints are uniformly distributed around the object by using the Fibonacci lattice

Download Views Generator


[S] J. Schöning, T. Behrens, P. Faion, P. Kheiri, G. Heidemann & U. Krumnack.
Structure from Motion by Artificial Neural Networks.
In Scandinavian Conference on Image Analysis (SCIA), pages: 146-158, ISBN: 978-3-319-59126-1, 2017. Springer International Publishing.
| PDF | DOI | URL | BibTeX