Slide 1

Slide 1 text

Beamerkalibrierungsuntersuchungstestvortragsfolie Quadrat Auflösung: 1920x1080 Stern in den Ecken zu sehen? Sowas wie 16:9? Nein? Soweit wie man es von Beamern erwarten kann? Quadrat quadratisch? Graustufen korrekt?

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

„Talk is cheap - Show me the video“ - Linus Torvalds, Mastermind behind Linux

Slide 4

Slide 4 text

Demo

Slide 5

Slide 5 text

No content

Slide 6

Slide 6 text

No content

Slide 7

Slide 7 text

No content

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

No content

Slide 11

Slide 11 text

No content

Slide 12

Slide 12 text

No content

Slide 13

Slide 13 text

No content

Slide 14

Slide 14 text

Video Propagation Networks Marcel Neidinger [email protected] 23/03/2018

Slide 15

Slide 15 text

Introduction Related Work and Background Video Propagation Networks Experimental Evaluation

Slide 16

Slide 16 text

Introduction Related Work and Background Video Propagation Networks Experimental Evaluation 00

Slide 17

Slide 17 text

Introduction

Slide 18

Slide 18 text

So what is the problem? 01

Slide 19

Slide 19 text

And what can I do with it? 02

Slide 20

Slide 20 text

And what can I do with it? 02

Slide 21

Slide 21 text

And what can I do with it? 02

Slide 22

Slide 22 text

And what can I do with it? 02

Slide 23

Slide 23 text

And what can I do with it? 03

Slide 24

Slide 24 text

And what can I do with it? 03

Slide 25

Slide 25 text

Duck Sea Background And what can I do with it? 03

Slide 26

Slide 26 text

Duck Sea Background And what can I do with it? 03

Slide 27

Slide 27 text

And what can I do with it? 04

Slide 28

Slide 28 text

And what can I do with it? 04

Slide 29

Slide 29 text

And what can I do with it? 04

Slide 30

Slide 30 text

And what can I do with it? 04

Slide 31

Slide 31 text

Summary: Introduction Recap 05

Slide 32

Slide 32 text

Summary: Introduction Recap 05

Slide 33

Slide 33 text

Summary: Introduction Video Segmentation Recap 05

Slide 34

Slide 34 text

Summary: Introduction Video Segmentation Recap Semantic Labeling 05

Slide 35

Slide 35 text

Summary: Introduction Video Segmentation Recap Semantic Labeling Color Propagation 05

Slide 36

Slide 36 text

Related Work and Background

Slide 37

Slide 37 text

General Propagation: Graphs 06

Slide 38

Slide 38 text

General Propagation: Graphs 06

Slide 39

Slide 39 text

General Propagation: Graphs 06

Slide 40

Slide 40 text

General Propagation: Graphs These Graphs can be Huge 07

Slide 41

Slide 41 text

General Propagation: Filtering 08

Slide 42

Slide 42 text

General Propagation: Filtering 08

Slide 43

Slide 43 text

General Propagation: Filtering 08

Slide 44

Slide 44 text

General Propagation: Filtering 08

Slide 45

Slide 45 text

Background: Bilateral Filtering Edge Preserving 09

Slide 46

Slide 46 text

Background: Bilateral Filtering Edge Preserving 09

Slide 47

Slide 47 text

Background: Bilateral Filtering 10

Slide 48

Slide 48 text

Background: Bilateral Filtering 10

Slide 49

Slide 49 text

Background: Bilateral Filtering ˆ vi = j∈n Wi,jvj 11

Slide 50

Slide 50 text

Background: Bilateral Filtering ˆ vi = j∈n Wi,jvj 11

Slide 51

Slide 51 text

Background: Bilateral Filtering ˆ vi = j∈n Wi,jvj g 11

Slide 52

Slide 52 text

Background: Bilateral Filtering x y 12

Slide 53

Slide 53 text

Background: Bilateral Filtering x y T 12

Slide 54

Slide 54 text

Bilateral Filtering: Splatting 13

Slide 55

Slide 55 text

Bilateral Filtering: Convolving B 14

Slide 56

Slide 56 text

Bilateral Filtering: slicing 15

Slide 57

Slide 57 text

Summary: Related Work Recap x y T B 16

Slide 58

Slide 58 text

Video Propagation Networks

Slide 59

Slide 59 text

The Task: A „formal“ Definition Black Box (VPN) 17

Slide 60

Slide 60 text

Architecture BNN CNN Bilateral Network Spatial Network 18

Slide 61

Slide 61 text

Bilateral Network: Bilateral Convolution Layer (BCL) 19

Slide 62

Slide 62 text

Bilateral Network: Bilateral Convolution Layer (BCL) 19

Slide 63

Slide 63 text

Bilateral Network: Bilateral Convolution Layer (BCL) 19

Slide 64

Slide 64 text

Bilateral Network: Bilateral Convolution Layer (BCL) x y T t 19

Slide 65

Slide 65 text

Bilateral Network: Bilateral Convolution Layer (BCL) x y T t 19

Slide 66

Slide 66 text

Bilateral Network: Bilateral Convolution Layer (BCL) x y T t 19

Slide 67

Slide 67 text

Bilateral Network: Bilateral Convolution Layer (BCL) B x y T t 19

Slide 68

Slide 68 text

Bilateral Network: Bilateral Convolution Layer (BCL) B x y T t 19

Slide 69

Slide 69 text

Bilateral Network: Bilateral Convolution Layer (BCL) B x y T t 19

Slide 70

Slide 70 text

Bilateral Network: Architecture Guidance BCLa BCLb || BCLa BCLb || C-1 C-3 C-3 C-3 || C-1 20

Slide 71

Slide 71 text

Summary: Video Propagation Networks Recap BNN CNN Bilateral Network Spatial Network 21 x y T t

Slide 72

Slide 72 text

Experimental Evaluation

Slide 73

Slide 73 text

Experiment: Video Object Segmentation 22

Slide 74

Slide 74 text

Experiment: Video Object Segmentation 22

Slide 75

Slide 75 text

Experiment: Video Object Segmentation 22

Slide 76

Slide 76 text

Video Object Segmentation: Dataset Densly Annotated VIdeo Segmentation 23

Slide 77

Slide 77 text

Video Object Segmentation: Evaluation 0 20 40 60 80 SEA JMP BNN-Identity VPN Stage 2 VPN-Deeplab IoU F T 24

Slide 78

Slide 78 text

Video Object Segmentation: Improvements 25

Slide 79

Slide 79 text

Video Object Segmentation: Improvements 68 67,5 67 66,5 66 0 500 1000 IoU Score Number of Points (in Thousand) 26

Slide 80

Slide 80 text

Video Object Segmentation: Improvements 12.000 SLIC Superpixel 27

Slide 81

Slide 81 text

Summary: Experimental Evaluation Recap -to- 28

Slide 82

Slide 82 text

Conclusion Three things to take away (i) VPNs use a bilateral and a spatial network (ii) Offer favorable runtime over current methods (iii) Perform better then current methods 1 1 And yes, this is the only textual slide you will see in this presentation

Slide 83

Slide 83 text

„The important thing is to never stop questioning“ - Albert Einstein, Mastermind of the Universe

Slide 84

Slide 84 text

Appendix • Definition of Performance Measures • Detailed Performance (Video Segmentation) • Detailed Performance (Semantic Labeling) • Detailed Performance (Color Propagation) • Literature Back to Overview • JumpCut (JMP) • SeamSeg (SEA)

Slide 85

Slide 85 text

Performance Measures Intersection over Union (IoU) Groundtruth G, Output segmentation M Contour Accuracy F Temporal Stability T Back to Overview

Slide 86

Slide 86 text

Detailed Performance (Video Segmentation) Back to Overview

Slide 87

Slide 87 text

Detailed Performance (Semantic Labeling) Back to Overview

Slide 88

Slide 88 text

Detailed Performance (Color Propagation) Back to Overview

Slide 89

Slide 89 text

Literature (besides the main paper) [1] S. Paris and F. Durand. A fast approximation of the bilateralfilter using a signal processing approach. In European Conference on Computer Vision, pages 568–580. Springer, 2006. [2] V. Jampani, M. Kiefel, and P. V. Gehler. Learning sparse high dimensional filters: Image filtering, dense CRFs and bilateral neural networks. In Computer Vision and Pattern Recognition, IEEE Conference on, June 2016. Back to Overview

Slide 90

Slide 90 text

JumpCut (JMP) Back to Overview • Key idea: Foreground and background exhibit different motions • Calculate two Nearest-Neighbor Fields to label target frame • Calculate Silhouette Edges (using the NNFs)

Slide 91

Slide 91 text

SeamSeg (SEA) Back to Overview • Key idea: Seams(Connected Paths of low energy) Energy formulation allows for temporal label propagation