unfolding an indoor
play

Unfolding an Indoor Origami World David Fouhey, Abhinav Gupta, - PowerPoint PPT Presentation

Unfolding an Indoor Origami World David Fouhey, Abhinav Gupta, Martial Hebert 1 2 3 Local Evidence Hoiem et al. 2005, Saxena et al. 2005, Fouhey et al. 2013, etc. 4 Constraints 5 Constraints for Single Image 3D Local Smoothness Low


  1. Unfolding an Indoor Origami World David Fouhey, Abhinav Gupta, Martial Hebert 1

  2. 2

  3. 3

  4. Local Evidence Hoiem et al. 2005, Saxena et al. 2005, Fouhey et al. 2013, etc. 4

  5. Constraints 5

  6. Constraints for Single Image 3D Local Smoothness Low Level, Generic Hoiem et al. 2005, Saxena et al. 2005, 2008, Munoz et al., 2009, etc. 6

  7. Constraints for Single Image 3D Local Smoothness Low Level, Generic Hoiem et al. 2005, Saxena et al. 2005, 2008, Munoz et al., 2009, etc. 7

  8. Constraints for Single Image 3D Low Level, High Level, Generic Physical 8

  9. Constraints for Single Image 3D Local Smoothness Low Level, High Level, Generic Physical Coughlan and Yuille 2000, etc. 9

  10. Constraints for Single Image 3D Local Smoothness Low Level, High Level, Generic Physical Hedau et al. 2009, Del Pero et al., 2011, Wang et al., 2012, Schwing et al. 2012, etc. 10

  11. Constraints for Single Image 3D Local Smoothness Low Level, High Level, Generic Physical Lee et al. 2010, Xiao et al. 2012, Zhao et al. 2013, Schwing et al., 2013, etc. 11

  12. Constraints for Single Image 3D Low Level, High Level, Generic Physical 12

  13. Mid-level in the Past Huffman 71, Clowes 71, Kanade 80, 81 Sugihara 86, Malik 87, etc. 13

  14. Our Mid-Level Constraints 14

  15. This Work Input: Output: Single Image Discrete Scene Parse 15

  16. Overview Parameterization Formulation Experimental Results 16

  17. Overview Parameterization Formulation Experimental Results 17

  18. Parameterization 18

  19. Parameterization vp 2 vp 3 vp 1 VP Estimator from Hedau et al., 2009 19

  20. Parameterization Two VPs give grid cell 20

  21. Encoding Surface Normals 21

  22. Encoding Surface Normals 22

  23. Encoding Surface Normals 23

  24. Encoding Surface Normals x 1 ,…, x 400 x 401 ,…, x 800 x 801 ,…, x 1200 24

  25. Related Parameterizations vp 2 x 4 vp 3 vp 1 x 2 x 3 x 1 Hedau et al., 2009; Wang et al. 2010, Schwing et al., 2012, 2013 25

  26. Overview Parameterization Formulation Experimental Results 26

  27. Parameterization 27

  28. Formulation 28

  29. Unaries 29

  30. Unaries Any High c Low c 3D Evidence 30

  31. Unaries Local: Data-Driven 3D Primitives Input 3D Primitive Bank … Fouhey, Gupta, Hebert, 2013 31

  32. Unaries Global: Cuboid Fit + Clutter Mask Input Predicted Walls Clutter Mask Hedau, Hoiem, Forsyth, 2009 32

  33. Binaries 33

  34. Convex/Concave Constraints Convex (+) Concave (-) 34

  35. Convex/Concave Constraints Detected Concave (-) 35

  36. Convex/Concave Constraints Detected Concave (-) 36

  37. Convex/Concave Constraints Detected Concave (-) 37

  38. Convex/Concave Constraints Detected Concave (-) 38

  39. Convex/Concave Constraints Detected Concave (-) 39

  40. Detecting Convex/Concave Use 3DP to Transfer Discontinuities Input 3D Primitive Bank … Ground-Truth Discontinuities similar to Gupta, Arbelaez, Malik, 2013 3DP from Fouhey, Gupta, Hebert, 2013 40

  41. Smoothness 41

  42. Constraints 42

  43. Solving the Model 43

  44. Overview Parameterization Formulation Experimental Results 44

  45. Dataset NYU Depth v2: 795 Train, 654 Test 45

  46. Qualitative Results 46

  47. Qualitative Results 47

  48. Qualitative Results 48

  49. Qualitative Results 49

  50. Surface Connection Graphs Convex Concave 50

  51. Baseline Primary Baseline: 3D Primitives Input Output Fouhey, Gupta, Hebert, 2013 51

  52. Quantitative Results % Good Pixels Summary Stats ( ⁰) (Lower Better) (Higher Better) Mean Median RMSE 11.25⁰ 22.5⁰ 30⁰ Proposed 35.1 19.2 48.7 37.6 53.3 58.9 3DP 36.0 20.5 49.4 35.9 52.0 57.8 Hedau et al. 40.0 23.5 54.1 34.2 49.3 54.4 Lee et al. 43.3 36.3 54.6 18.6 38.6 49.9 52

  53. Quantitative Results 53

  54. Failure Modes Mistaken but Confident Evidence 54

  55. Failure Modes Missing High-Level Modeling 55

  56. Conclusions Parameterization Discrete Parse Single Image Formulation 56

  57. Thank You 57

Recommend


More recommend