Structured Forests for ConBnuity Fast Edge Detection Symmetry - - PowerPoint PPT Presentation

structured forests for
SMART_READER_LITE
LIVE PREVIEW

Structured Forests for ConBnuity Fast Edge Detection Symmetry - - PowerPoint PPT Presentation

4/27/16 what defines an edge? Brightness Color Texture Parallelism Structured Forests for ConBnuity Fast Edge Detection Symmetry Piotr Dollr and Larry Zitnick 1. A 1. Accur urac acy y 2. Speed 2. Speed I. data driven edge


slide-1
SLIDE 1

4/27/16 1

Structured Forests for Fast Edge Detection

Piotr Dollár and Larry Zitnick

what defines an edge?

Brightness Color Texture Parallelism ConBnuity Symmetry …

  • 1. A
  • 1. Accur

urac acy y

  • 2. Speed
  • 2. Speed
  • I. data driven edge detec;on
slide-2
SLIDE 2

4/27/16 2

edge detec;on as classifica;on

Supervised Learning of Edges and Object Boundaries CVPR 2006, Piotr Dollár, Zhuowen Tu, Serge Belongie

posiBves

{ 0, 1 }

edge have structure sketch tokens

Sketch Tokens, CVPR 2013. Joseph Lim, C. Zitnick, and P. Dollár

random forests

slide-3
SLIDE 3

4/27/16 3

upgrading the output space

{ 0, 1 } { … }

dimensionality

2 2256 151

  • II. structured edge learning

structured forests

Structured Class-Labels in Random Forests for SemanBc Image Labelling, ICCV 2011,

  • P. Kontschieder, S. Rota Bulò, H. Bischof, M. Pelillo

tree training

slide-4
SLIDE 4

4/27/16 4

node training

high entropy split L low entropy split J

how to train?

bad split L good split J ?

? ? ?

cluster minimize entropy

  • III. structured edge detec;on
slide-5
SLIDE 5

4/27/16 5

structured forests sliding window detector sliding window detector

pixel output L structured output J

mul;scale detec;on

slide-6
SLIDE 6

4/27/16 6

½ x 2 x 1 x

mul;scale detec;on

+ +

  • IV. results

SS=single-scale MS=mul8-scale T=# trees 6Hz SS T=1 SS T=4 MS T=4 ODS = 0.72 ODS = 0.74 ODS = 0.73 60Hz 30Hz gPb ODS=.73 gPb ODS=.73 FPS ≈ 1/240 Hz

slide-7
SLIDE 7

4/27/16 7

thanks!

source code available online