NMTA Economic Forecast November 9, 2011 Jim Hebert Hebert Research, Inc. 1.
Unfolding an Indoor Origami World David Fouhey, Abhinav Gupta, Martial Hebert 1.
-
Upload
lynne-wiggins -
Category
Documents
-
view
225 -
download
5
Transcript of Unfolding an Indoor Origami World David Fouhey, Abhinav Gupta, Martial Hebert 1.
1
Unfolding an Indoor Origami World
David Fouhey, Abhinav Gupta, Martial Hebert
2
3
4
Local Evidence
Hoiem et al. 2005, Saxena et al. 2005, Fouhey et al. 2013, etc.
5
Constraints
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.
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.
8
Constraints for Single Image 3D
Low Level,Generic
High Level,Physical
9
High Level,Physical
Constraints for Single Image 3D
Local Smoothness
Low Level,Generic
Coughlan and Yuille 2000, etc.
10
High Level,Physical
Constraints for Single Image 3D
Local Smoothness
Low Level,Generic
Hedau et al. 2009, Del Pero et al., 2011, Wang et al., 2012, Schwing et al. 2012, etc.
11
High Level,Physical
Constraints for Single Image 3D
Local Smoothness
Low Level,GenericLee et al. 2010, Xiao et al. 2012, Zhao et al. 2013, Schwing et al., 2013,
etc.
12
High Level,Physical
Constraints for Single Image 3D
Low Level,Generic
13
Mid-level in the Past
Huffman 71, Clowes 71, Kanade 80, 81 Sugihara 86, Malik 87, etc.
14
Our Mid-Level Constraints
15
This WorkInput:
Single ImageOutput:
Discrete Scene Parse
16
Overview
Parameterization
Formulation
Experimental Results
17
Overview
Parameterization
Formulation
Experimental Results
18
Parameterization
19
Parameterization
vp1
vp2
vp3
VP Estimator from Hedau et al., 2009
20
Parameterization
Two VPs give grid cell
21
Encoding Surface Normals
22
Encoding Surface Normals
23
Encoding Surface Normals
24
Encoding Surface Normals
x1,…, x400 x401,…, x800
x801,…, x1200
25
Related Parameterizations
vp1
vp3
x2
x1
x3
x4
Hedau et al., 2009; Wang et al. 2010, Schwing et al., 2012, 2013
vp2
26
Overview
Parameterization
Formulation
Experimental Results
27
Parameterization
28
Formulation
29
Unaries
30
Unaries
Low cAny 3D Evidence
High c
31
Unaries
…
Input 3D Primitive Bank
Local: Data-Driven 3D Primitives
Fouhey, Gupta, Hebert, 2013
32
Unaries
Hedau, Hoiem, Forsyth, 2009
Global: Cuboid Fit + Clutter Mask
Input Predicted Walls Clutter Mask
33
Binaries
34
Convex/Concave Constraints
Convex (+) Concave (-)
35
Convex/Concave Constraints
Detected Concave (-)
36
Convex/Concave Constraints
Detected Concave (-)
37
Convex/Concave Constraints
Detected Concave (-)
38
Convex/Concave Constraints
Detected Concave (-)
39
Convex/Concave Constraints
Detected Concave (-)
40
Detecting Convex/Concave
Ground-Truth Discontinuities similar to Gupta, Arbelaez, Malik, 2013
3DP from Fouhey, Gupta, Hebert, 2013
Input 3D Primitive Bank
…
Use 3DP to Transfer Discontinuities
41
Smoothness
42
Constraints
43
Solving the Model
44
Overview
Parameterization
Formulation
Experimental Results
45
Dataset
NYU Depth v2: 795 Train, 654 Test
46
Qualitative Results
47
Qualitative Results
48
Qualitative Results
49
Qualitative Results
50
Surface Connection Graphs
Convex Concave
51
Baseline
Primary Baseline: 3D Primitives (3DP)
Fouhey, Gupta, Hebert, 2013
OutputInput
52
Quantitative Results
Summary Stats (⁰)
(Lower Better)
% Good Pixels(Higher Better)11.2
5⁰22.5
⁰30⁰Mea
nMedia
n
3DP 35.9
52.0
57.8
36.0
20.5 49.4
Hedau et al.
34.2
49.3
54.4
40.0
23.5 54.1
RMSE
Lee et al. 18.6
38.6
49.9
43.3
36.3 54.6
Proposed 37.6
53.3
58.9
35.1
19.2
48.7
53
Quantitative Results
54
Failure Modes
Mistaken but Confident Evidence
55
Failure Modes
Missing High-Level Modeling
56
Conclusions
Single Image
Parameterization
Formulation
Discrete Parse
57
Thank You