Deep Learning based Aesthetic Evaluation of State-Of-The-Art 3D Reconstruction Techniques - 17.306

G. Stuebl et al., "Deep Learning based Aesthetic Evaluation of State-Of-The-Art 3D Reconstruction Techniques", in Proc. of 3DBODY.TECH 2017 - 8th Int. Conf. and Exh. on 3D Body Scanning and Processing Technologies, Montreal QC, Canada, 11-12 Oct. 2017, pp. 306-311, https://doi.org/10.15221/17.306.

Title:

Deep Learning based Aesthetic Evaluation of State-Of-The-Art 3D Reconstruction Techniques

Authors:

Gernot STUEBL, Christoph HEINDL, Harald BAUER, Andreas PICHLER

PROFACTOR GmbH, Steyr-Gleink, Austria

Abstract:

In the field of 3D human body scanning, due to different scanning technologies different reconstruction approaches have emerged. The two main ones are based either on pure 2D information, like photogrammetry, or 2D plus depth, as used with RGBD active structured light sensors. Reconstruction results of these technologies differ in geometric as well as aesthetic quality. Whereas the judgement of geometric quality is straight forward, a judgement of the aesthetics aspects (e.g. proper texture mapping, etc.) strongly depends on the subjective perception of the human viewer. Recent advances in image aesthetics assessment, demonstrate that machine-learning algorithms, specifically deep neural networks, are able to model human aesthetic perception in a reasonable way. Especially if they are trained with a huge number of data. This work presents research towards an unbiased aesthetic judgement of 3D reconstructions by utilizing a deep neural network. In detail, two state-of-the-art software suites as representatives for 2D and 2D plus depth reconstruction approaches are compared according to the aesthetics of their results. The models of a publicly available dataset are virtually scanned with a sensor simulator, which produces the necessary 2D and depth information. This data serves as input to the mentioned software suites. The resulting 3D reconstructions are aligned and a deep neural net aesthetically compares frontal views of the models. To ensure a fair comparison between different models a normalized aesthetic value is introduced.

Details:

Full paper: 17.306.pdf
Proceedings: 3DBODY.TECH 2017, 11-12 Oct. 2017, Montreal QC, Canada
Pages: 306-311
DOI: 10.15221/17.306

License/Copyright notice:

Proceedings: © Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Authors retain all rights to individual papers, which are licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
The papers appearing in the proceedings reflect the author's opinions. Their inclusion in the proceedings does not necessary constitute endorsement by the editor or by the publisher.


Note: click the + on the top left of the page to open/close the menu.