1 / 39

3D Model Retrieval

3D Model Retrieval. After Shape Distributions. Agenda. Shape Distributions and selected descendants. Shape Distributions S.D. on Solid Models AD & AAD Princeton Shape Benchmark Statistics Benchmark Statistical tools to our system. Papers.

Download Presentation

3D Model Retrieval

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 3D Model Retrieval After Shape Distributions Sheun-Huei Guan, CML, NTU

  2. Agenda • Shape Distributions and selected descendants. • Shape Distributions • S.D. on Solid Models • AD & AAD • Princeton Shape Benchmark • Statistics • Benchmark • Statistical tools to our system Sheun-Huei Guan, CML, NTU

  3. Papers • R. Osada, T. Funkhouser, B. Chazelle, D. Dobkin, Shape Distributions, ACM ToG 2002. • C. Ip, D. Lapadat, L. Sieger, W. Regli, Using Shape Distributions to Compare Solid Models, 7th ACM/SIGGRAPH Symp. on Solid Modeling and Applications 2002. • R. Ohbuchi, T. Minamitani, T. Takei, Shape-Similarity Search of 3D Models by using Enhanced Shape Functions, Theory and Practice of Computer Graphics 2003. • P. Shilane, P. Min, M. Kazhdan, T. Funkhouser, The Princeton Shape Benchmark, Shape Modeling International 2004. Sheun-Huei Guan, CML, NTU

  4. Shape Distributions Sheun-Huei Guan, CML, NTU

  5. Shape Distributions Sheun-Huei Guan, CML, NTU

  6. Shape Functions Sheun-Huei Guan, CML, NTU

  7. Shape Function: D2 • Sample points on mesh uniformly • For each point p, get D2 (histogram) • Sum all D2 (histogram) • Transform the histogram to coefficients Sheun-Huei Guan, CML, NTU

  8. Shape Distributions Sheun-Huei Guan, CML, NTU

  9. Comparing Shape Distributions • Bhattacharyya: D(f, g) = 1 – • PDF LN: Minkowski LN norm • L1 norm • L2 norm • … Sheun-Huei Guan, CML, NTU

  10. Issues • Point sampling! • Gross shape properties! Sheun-Huei Guan, CML, NTU

  11. Issue: point sampling • Pseudo Random Number Generator?! Sheun-Huei Guan, CML, NTU

  12. Issue: gross shape properties Sheun-Huei Guan, CML, NTU

  13. Solid Model Sheun-Huei Guan, CML, NTU

  14. Solid Model • Gross D2 on surface is not enough. • Accumulate Distribution of Classification Sheun-Huei Guan, CML, NTU

  15. Histogram (IN, OUT, MIXED) Sheun-Huei Guan, CML, NTU

  16. Only D2? Sheun-Huei Guan, CML, NTU

  17. Enhanced Shape Functions • Angle and Distance (AD) & Absolute Angle and Distance (ADD). Sheun-Huei Guan, CML, NTU

  18. AD & AAD • AD/AAD is a 2D histogram Sheun-Huei Guan, CML, NTU

  19. Results with AD & AAD Sheun-Huei Guan, CML, NTU

  20. Statistical Analysis Methods Sheun-Huei Guan, CML, NTU

  21. Statistics • Given a class with |C| members • The percentage of models in the query’s class that appear within the top K matches. • 1-tier: K = |C| - 1 • 2-tier: K = 2(|C| - 1) • nearest neighbor: K = 1 • Discounted Cumulative Gain (DCG) Sheun-Huei Guan, CML, NTU

  22. Precision / Recall • P := γ/ (γ+α) • R := γ/ (γ+β) Sheun-Huei Guan, CML, NTU

  23. Diagram Precision ideal 1 0 1 Recall Precision Recall 1 0 1 1/4 3/4 2/4 4/8 3/4 5/15 4/4 Sheun-Huei Guan, CML, NTU

  24. Princeton Shape Benchmark Sheun-Huei Guan, CML, NTU

  25. What is PSB • Providing a standard 3D model database. • Classifying models by semantic, function and shape. • Doing benchmark. Sheun-Huei Guan, CML, NTU

  26. Results: Precision-Recall Sheun-Huei Guan, CML, NTU

  27. Results: Statistics Sheun-Huei Guan, CML, NTU

  28. Results: Different Classification Sheun-Huei Guan, CML, NTU

  29. 3D Model Retrieval byLight Field Sheun-Huei Guan, CML, NTU

  30. 3 4 1 2 3 4 5 1 2 4 5 1 2 3 4 5 1 2 3 5 …… …… Matching (I) Sheun-Huei Guan, CML, NTU

  31. 1 5 5 1 2 3 4 1 2 3 4 4 5 3 3 2 1 2 4 5 2 1 5 4 3 …… …… Matching (II) …… Sheun-Huei Guan, CML, NTU

  32. 1 5 1 2 3 4 5 1 2 3 4 4 5 3 3 2 1 2 4 3 1 2 5 4 5 …… …… Matching (III) …… Sheun-Huei Guan, CML, NTU

  33. 2 5 5 4 3 1 2 1 2 3 5 1 4 3 3 2 1 4 4 5 2 1 5 4 3 …… …… Matching (IV) …… Sheun-Huei Guan, CML, NTU Minimum error from all corresponding 2D shapes

  34. To be more robust in rotation(I) Sheun-Huei Guan, CML, NTU

  35. 1-10 600 To be more robust in rotation(II) 1-1 60 10-10 5460 Different rotation: (Mx(N-1)+1)x60 Sheun-Huei Guan, CML, NTU

  36. Matching as Histogram 1 Number 0 0 1 Similarity Sheun-Huei Guan, CML, NTU

  37. Q1 Median Q3 Matching as Histogram with IQR 1 Number 0 0 1 Similarity Sheun-Huei Guan, CML, NTU

  38. Results with IQR Sheun-Huei Guan, CML, NTU

  39. Thank You See you in the next retrieval. Sheun-Huei Guan, CML, NTU

More Related