SHARP - A Benchmark for Automatic Best View Selection of 3D Objects

BEST VIEW SELECTION corresponds to the task of automatically selecting the most representative view of a 3D model. 

This benchmark aims to provide tools to evaluate automatic best view selection algorithms. It consists of the preferred viewpoints of 68 models selected by 26 human subjects, and a code to calculate "View selection error".

The subjective viewpoints were collected using a web-based subjective experiment where the users were asked to select the most informative view of a 3D model.

Please refer to the paper for more information about this benchmark:  Helin Dutagaci, Chun Pan Cheung, Afzal Godil "A benchmark for best view selection of 3D objects <filename:  3dor3737-dutagaci.pdf>" 3DOR '10 Proceedings of the ACM workshop on 3D object retrieval [Bib Tex <filename: bibtex.bib>]

POSTER PRESENTATION <filename:  3DOR-BestView-Poster.pdf>

3D Model Data Set

The 3D object dataset consists of 68 triangular meshes. Some of the models are standard models that are widely used in 3D shape research; and they have been used as test objects by researchers working on the best view problem. Examples are Armadillo, David’s head, Utah teapot, Bunny, etc. We chose some of the models from The Stanford 3D Scanning Repository and some others from the Watertight Models Track of SHREC 2007 (see the license <filename:  AIM.txt>).

Download the DATA SET <filename:  EXP_MODELS.zip>. The data set is also available in lower resolution: DATA SET (LR) <filename:  EXP_MODELS_LR.zip>

Web-based interface

Users can login to our web-based interface using an alias and participate in the experiments. The user is shown the 68 3D models one at a time. Each model is initially rendered with a random pose. The user is asked to rotate the model via dragging the mouse into a view that he/she thinks is the best, and then to click on the submit usa-button.

Bestview Interface Demo

Subjective viewpoints

26 participants have submitted their preferred best views for the 68 models. For each model, we provide an ASCII file containing the view-points selected by the human subjects. 
Download the SUBJECTIVE VIEWPOINT LISTS <filename:  SUBJECTS_VIEWPOINTS.zip>of the 68 models (README <filename:  README-subjects-viewpoints.txt>for the file format). 

Symmetry sets

Our evaluations take into account the symmetry sets of the viewpoints. We have used the Fourier-Mellin image matching technique to determine symmetries in the object with respect to a particular view-point.

We provide a symmetry-set file for each model. We calculated the symmetry sets on the 258 vertices of a unit geodesic sphere surrounding the model. The x y z coordinates of the vertices are in the ASCII file geodesic_sphere.txt <filename: geodesic_sphere.txt>. The index of each vertex is its row order in the file.

Download the SYMMETRY SET LISTS <filename:  SYMMETRY_SETS.zip> of the 68 models (README <filename:  README-symmetry-sets.txt> for the file format).

Evaluation Measure

For a specific 3D model, given the best viewpoint generated by an algorithm, we can evaluate the "goodness" of that viewpoint by the "View Selection Error". This evaluation easure makes use of the subjective viewpoints, the 258 point geodesic sphere, and the symmetry sets of the model. 

Download the MATLAB code <filename:  evaluation_code.zip>for calculating the View Selection Error.

Best-view selection algorithms

We have compared the following best view selection algorithms: 
"Matrix of Best Viewpoints" is a MATLAB data file containing the best viewpoints of the 68 models calculated by the corresponding algorithm.  It is a 68x3 matrix, and each row gives the x y z coordinates of the best view of the model, calculated by the algorithm. The viewpoints are in the same order of the model names in the model list file "model_list.txt" <filename: model_list.txt >.

• View area [Matrix of Best Viewpoints] <filename:  area_BV_matrix.zip> 
• Ratio of visible area [Polonsky et al. 2005] [Matrix of Best Viewpoints] <filename:  visible_area_ratio_BV_matrix.zip> 
• Surface area entropy [Vázquez et al. 2001] [Matrix of Best Viewpoints] <filename:  area_entropy_BV_matrix.zip> 
• Silhouette length [Polonsky et al. 2005] [Matrix of Best Viewpoints] <filename:  silhouette_length_BV_matrix.zip> 
• Silhouette entropy [Page et al. 2003] [Matrix of Best Viewpoints] <filename:  silhouette_entropy_BV_matrix.zip> 
• Curvature entropy [Page et al. 2003] [Matrix of Best Viewpoints] <filename:  curvature_entropy_BV_matrix.zip> 
• Mesh saliency [Lee et al. 2005] [Matrix of Best Viewpoints] <filename:  mesh_saliency_BV_matrix.zip> 

ALL MATRICES OF BEST VIEWPOINTS IN A ZIP FILE <filename:  MATRICES_OF_BEST_VIEWS.zip> 

The MATLAB implementation of these methods is now available below; If any issue(s) please contact Helin Dutagaci or Afzal Godil.

Evaluation Results

 
See the Excel chart <filename:  BestViewEvaluation.xlsx>, which includes View Selection Errors on individual models

Downloads

DATA SET <filename:  EXP_MODELS.zip>The data set is also available in lower resolution: DATA SET (LR) <filename:  EXP_MODELS_LR.zip>
The VISIBLE vertices of the low resolution models <filename:  >
The Mesh Saliency of the low resolution models <filename:  >
SUBJECTIVE VIEWPOINT LISTS <filename:  SUBJECTS_VIEWPOINTS.zip> of the 68 models (README <filename:  README-subjects-viewpoints.txt> for the file format)
SYMMETRY SET LISTS <filename:  SYMMETRY_SETS.zip> of the 68 models (README <filename:  README-symmetry-sets.txt> for the file format) 
geodesic_sphere.txt <filename:  geodesic_sphere.txt>
model_list.txt <filename: model_list.txt>
ALL MATRICES OF BEST VIEWPOINTS <filename:  MATRICES_OF_BEST_VIEWS.zip>
MATLAB code <filename:  evaluation_code.zip> for calculating the View Selection Error.
Excel chart <filename:  BestViewEvaluation.xlsx>, which includes View Selection Errors on individual models.
The MATLAB implementation of these Best View methods <filename:  best_view_codes.rar>

EVERYTHING <filename:  BestViewBenchmarkNew.zip>

Citation

 Helin Dutagaci, Chun Pan Cheung, Afzal Godil "A benchmark for best view selection of 3D objects <filename:  3dor3737-dutagaci.pdf>" 3DOR '10 Proceedings of the ACM workshop on 3D object retrieval [Bib Tex <filename: bibtex.bib>]

References

• C. H. Lee, A. Varshney, and D. W. Jacobs. "Mesh saliency". ACM Trans. Graph., 24(3):659-666, 2005.
• D. L. Page, A. Koschan, S. R. Sukumar, B. Roui-Abidi, and M. A. Abidi. "Shape analysis algorithm based on information theory". In ICIP (1), pages 229-232, 2003.
• O. Polonsky, G. Patanµe, S. Biasotti, C. Gotsman, and M. Spagnuolo. "What's in an image? ", The Visual Computer, 21(8-10):840-847, 2005.
• P.-P. Vázquez, M. Feixas, M. Sbert, and W. Heidrich. "Viewpoint selection using viewpoint entropy". In VMV'01, pages 273-280. Aka GmbH, 2001.

Contact Information

For more information and feedback, please contact Helin Dutagaci or Afzal Godil .