The 2D shape structure dataset: A user annotated open access - PowerPoint PPT Presentation

The 2D shape structure dataset: A user annotated open access database A. Carlier, G. Morin K. Leonard S. Hahmann M. Collins IRIT – INP Toulouse CSUCI INRIA Grenoble GISHWHES 24-06-2016 SMI'16 1

Motivation [Liu et al.] Convex shape decomposition. In: IEEE conference on computer vision and pattern recognition (CVPR); 2010. [Lien et al.] Approximate convex decomposition of polygons. In: Comput Geom, 35 (1–2) (2006), pp. 100–123 [Luo et al.] A computational model of the short-cut rule for 2D shape decomposition. In: IEEE Trans Image Process, 24 (1) (2015), pp. 273–283 Ground Truth ? 24-06-2016 SMI'16 2

Shape structure Details Main Shape Parts 24-06-2016 SMI'16 3

Existing datasets 20 categories 10 shapes per category 12 human annotations per shape Segmentation, not hierarchy [Jiang et al.] Toward perception-based shape decomposition. In: Computer vision, ACCV. Springer; 2012. p. 188–201. 24-06-2016 SMI'16 4

Existing datasets 88 shapes 201 users 122 human annotations per shape in average Cuts were drawn, but no hierarchy was given [De Winter et al.] Segmentation of object outlines into parts: a large-scale integrative study. In: Cognition, 99 (3) (2006), pp. 275–325 24-06-2016 SMI'16 5

Setup ● 1253 shapes, 70 categories ● All shapes are from the MPEG7 database, plus a few artificial shapes ● Sub-sampling the boundary curves ● Delaunay triangulation (~100 triangles / shape) 24-06-2016 SMI'16 6

Setup 24-06-2016 SMI'16 7

Setup ● 15,000 participants, grouped in teams of 5 ● Our task is one item, out of 212 items ● Annotate 20 shapes 24-06-2016 SMI'16 8

Setup ● Tutorial ● 4 Gold Standard shapes 24-06-2016 SMI'16 9

Collected data ● 2,861 teams started the task, 1,877 completed it ● 41,953 annotated shapes ● At least 24 annotations per shape 24-06-2016 SMI'16 10

Collected data 24-06-2016 SMI'16 11

Collected data 24-06-2016 SMI'16 12

Collected data 24-06-2016 SMI'16 13

Errors Random mistakes Bad understanding of task Spamming Oleson, D., Sorokin, A., Laughlin, G. P., Hester, V., Le, J., & Biewald, L. (2011). Programmatic Gold: Targeted and Scalable Quality Assurance in Crowdsourcing. Human computation, 11(11). 24-06-2016 SMI'16 14

Filtering users Distance Quality of user u 24-06-2016 SMI'16 15

Worst users 24-06-2016 SMI'16 16

Majority Vote 24-06-2016 SMI'16 17

Is the majority vote satisfying? 24-06-2016 SMI'16 18

Is the majority vote satisfying? 24-06-2016 SMI'16 19

Spectral clustering Affinity Matrix Ng, A. Y., Jordan, M. I., & Weiss, Y. (2002). On spectral clustering: Analysis and an algorithm. Advances in neural information processing systems, 2, 849-856. 24-06-2016 SMI'16 20

Results 24-06-2016 SMI'16 21

Conclusion ● Dataset is available at: http://2dshapesstructure.github.io/index.html ● Raw data is presented along with processed results: majority vote, and spectral clustering ● All related code can be found here: https://github.com/2DShapesStructure/ 24-06-2016 SMI'16 22