03 - Detecting Single Objects in Context

download 03 - Detecting Single Objects in Context

of 34

Transcript of 03 - Detecting Single Objects in Context

  • 8/8/2019 03 - Detecting Single Objects in Context

    1/34

    C o n text fo r sin g leo b je ct cla sse s

  • 8/8/2019 03 - Detecting Single Objects in Context

    2/34

    o nee s con ex anywayWe can recognize objects even out of

    context

    Banksy

  • 8/8/2019 03 - Detecting Single Objects in Context

    3/34

    Why is context important?Changes the interpretation of an object (or its function)

    Context defines what an unexpected event is

  • 8/8/2019 03 - Detecting Single Objects in Context

    4/34

    Look-Alikes by Joan Steiner

    ,Even in high resolution we can not shut down contextual processing and

    it is hard to recognize the true identities of the elements that compose.this scene

  • 8/8/2019 03 - Detecting Single Objects in Context

    5/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    6/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    7/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    8/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    9/34

    The importance of context

    Cognitive psychology Palmer 1975

    Biederman 1981

    Computer vision Noton and Stark (1971)

    Hanson and Riseman (1978)

    Barrow & Tenenbaum (1978) Ohta, kanade, Skai (1978)

    Haralick (1983)

    Strat and Fischler (1991)

    Bobick and Pinhanez (1995)

    Campbell et al (1997)

  • 8/8/2019 03 - Detecting Single Objects in Context

    10/34

    What is the context for asingle object category?

  • 8/8/2019 03 - Detecting Single Objects in Context

    11/34

    The influence of an objectextends beyond its physical

    boundaries

  • 8/8/2019 03 - Detecting Single Objects in Context

    12/34

    representations

    building

    car

    sidewalk

    Urban streetscene

  • 8/8/2019 03 - Detecting Single Objects in Context

    13/34

    representations

    : ,Image index Summary statisticsconfiguration of textures

    Urban streetscene

    features

    histogram

    building

    car

    sidewalk

    Urban streetscene

  • 8/8/2019 03 - Detecting Single Objects in Context

    14/34

    Global scenerepresentations

    Spatial structure is important in order to provide context forobject localization

    . .,Sivic et al ICCV 2005- ,Fei Fei and Perona CVPR 2005

    Bag of words Spatially organized textures

    Non localized textons

    . , ,S Lazebnik et al CVPR 2006

    , . Walker Malik Vision Research2004

    . , . ,M Gorkani R Picard ICPR 1994

    . , . ,A Oliva A Torralba IJCV 2001

  • 8/8/2019 03 - Detecting Single Objects in Context

    15/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    16/34

    Context driven object detection

    ZcarNcar

    S

    g

    Scene

    Scenegist

    features

    0 1 5

    (P Ncar = )| S street

    0

    0.05

    0.1

    0.15

    0.2

    N

  • 8/8/2019 03 - Detecting Single Objects in Context

    17/34

    ,An integrated model of Scenes,Objects and Parts

    ( = ) =p d | F 1 N(d | 1, 1)

    ( = ) =p d | F 0 N(d | 0, 0)

    . We train a multiview car detector

    xcar

    i

    dcar

    i

    carF

    i

    =N 4

  • 8/8/2019 03 - Detecting Single Objects in Context

    18/34

    ,An integrated model of Scenes,Objects and Parts

    ZcarNcar

    S

    g

    Scene

    Scenegist

    features

    xcar

    i

    dcar

    i

    carFi

    =M 4

    ( , , , )F S | x d g ( ) ( ) (p F | S p S | g p xi )| g N(xi; b, b2) N(di; tp , tp 2)N(di; tn , tn

    2) :i Fi=0 :i Fi=1 :i Fi=0

  • 8/8/2019 03 - Detecting Single Objects in Context

    19/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    20/34

  • 8/8/2019 03 - Detecting Single Objects in Context

    21/34

    A car out of context

  • 8/8/2019 03 - Detecting Single Objects in Context

    22/34

    W e a re w ire d fo r

    3 D ~6cm

    We can not shut do n 3D

  • 8/8/2019 03 - Detecting Single Objects in Context

    23/34

    We can not shut down 3Dperception

    ( )c 2006 Walt Anthony

  • 8/8/2019 03 - Detecting Single Objects in Context

    24/34

    Scenes rule over objects

    ,3 D p e rce p t is d rive n b y th e sce n e w h ich im p o se s its ru lin g to th e

    o b je cts

  • 8/8/2019 03 - Detecting Single Objects in Context

    25/34

    3D from pixel values

    . , . . , . , " - . .D Hoiem A A Efros and M Hebert Automatic Photo Pop up SIGGRAPH 2005

    . , . , . . . " -A Saxena M Sun A Y Ng Learning 3 D Scene Structure from a Single Still Imag( - ), .In ICCV workshop on 3D Representation for Recognition 3dRR 07 2007

  • 8/8/2019 03 - Detecting Single Objects in Context

    26/34

    Surface EstimationImage Support Vertical Sky

    -V Left -V Center -V Right -V Porous -V Solid

    [ , , ]Hoiem Efros Hebert ICCV 2005

    ObjectSurface?

    Support? Slide by Derek Hoiem

  • 8/8/2019 03 - Detecting Single Objects in Context

    27/34

    Object Support

    Slide by Derek Hoiem

  • 8/8/2019 03 - Detecting Single Objects in Context

    28/34

    Slide by James Coughlan

  • 8/8/2019 03 - Detecting Single Objects in Context

    29/34

    y J ug

  • 8/8/2019 03 - Detecting Single Objects in Context

    30/34

    3d Scene Context

    Image World

    , ,Hoiem Efros Hebert ICCV 2005

    3D t t

  • 8/8/2019 03 - Detecting Single Objects in Context

    31/34

    3D scene context

    meters

    me

    te

    rs

    Ped

    Ped

    Car

    , ,Hoiem Efros Hebert ICCV 2005

  • 8/8/2019 03 - Detecting Single Objects in Context

    32/34

    Qualitative Results

    : /Initial 2 TP 3FP

    : /Final 7 TP 4FP

    [ - - ]Local Detector from Murphy Torralba Freeman 2003

    :Car TP / FP :Ped TP / FP

    Slide by Derek Hoiem

  • 8/8/2019 03 - Detecting Single Objects in Context

    33/34

    3D City Modeling using CognitiveLoops

    . , . , . , . .N C orne lis B Leibe K C orne lis L Van G ool C V PR '06

  • 8/8/2019 03 - Detecting Single Objects in Context

    34/34

    Single view metrology, .C rim in isi e t a l 1 9 9 9

    :N e e d to re cov e r G ro u n d p la n e R e fe re n ce h e ig h t H o rizo n lin e W h e re o b je cts co n ta ct th e

    d