Introduction to YOLO detection model

  • Published on
    15-Apr-2017

  • View
    646

  • Download
    4

Embed Size (px)

Transcript

  • YOLO

    2017/1/13

  • [1] J Redmon, et al.You Only Look Once: Unified, Real-Time Object DetectionarXiv:1506.02640v5 9 May 2016. 1

  • Detection

    R-CNN

    (sliding window)CNN

  • Detection

  • l l

    S S

    S

    S

    B

    B = 3 0.44

    0.71 0.08

    Pr Object( ) IOUpredtruth

    = Pr Object( ) pred truthpred truth

    B1 B2

    B3

  • l

    x, y( )

    S

    S

    h,w( ) F

    B1 = x1, y1,h1,w1,F1( )x1, y1( ) h1w1

  • l C

    S

    S

    C = Pr Class |Object( )

    C1 = Pr Class |OO( ) = 0.14ex)

    C2 = Pr Class |OO( ) = 0.74C3 = Pr Class |OO( ) = 0.12

  • l 77 2

    l F

    B x, y,h,w,F( )

  • F

    B x, y,h,w,F( )

  • B x, y,h,w,F( )

    C = Pr Class |Object( )

  • l

  • convImageNet

    GoogLeNet

  • l conv l ReLuLeakly Relu

  • C1 C20

    x1 y1 w1h1 F1

    B1 B2

    7 7 C = 20B = 25B+C = 52+ 20 = 30

  • l sum squared loss

    L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    noobj = 0.5coord = 5

    ijobj = 1

    0

    iobj = 1

    0

    object i

    j i

  • l

    L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

  • l

    L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    L 0

    0

    0 0

    0 0

    0

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    0.2, 0.6( )

    0.1, 0.8( )

    0.7, 0.2( )

    l indicator )

    l l

    0.4, 0.1( )0.9, 0.2( )

    0.3, 0.9( )

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    l indicator )

    l

    l

    0.3, 0.4( )0.6, 0.4( )

    0.3, 0.1( ) 0.3, 0.2( )

    0.7, 0.6( )0.3, 0.4( )

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    l

    indicator )

    l

    11

    1

    0.2

    0.4

    0.7

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    l

    indicator )

    l

    0.2

    0.2

    0.4

    0.4

    0.4 0.4

    0.4 0.4

    0.4

    0.2

    0.2

    0.2

    0.1

    0.1

    0.1

    0.1

    0.3

    0.3

    0.3

    0.5

    0.5

    0.50.5

    0.6

    0.6

    0.6

    0.7

    0.7

    0.7

    0.7

    0.7

    0.8

    0.8

    0.8

    0.8

    0.9

    0.9

    0.9

    0.9

    0.6

    0.20.6

    0.6

    0

    1

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    l indicator )

    l

    l

  • L = coord ijobj xi xi( )

    2+ yi yi( )

    2

    j=0

    B

    i=0

    S2

    +coord ijobj wi wi( )2+ hi hi( )

    2

    j=0

    B

    i=0

    S2

    + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

    s s B indicator

    s s B indicator

  • + ijobj Ci Ci( )

    2

    j=0

    B

    i=0

    S2

    +noobj ijnoobj Ci Ci( )2

    j=0

    B

    i=0

    S2

    + iobj pi c( ) pi c( )( )2

    cclasses

    i=0

    S2

  • epoch 135 data sets PASCAL VOC 2007, 2012batch size 64optimizer momentum dropout

    data augmentation

  • l

  • l RCNNDPM

  • l YOLO

  • l YOLOCNN

    l