ó
|
×íZc @ sG d Z d d l m Z d d l Z d d l Z d d l m Z d d l
|
m Z m Z m Z m Z m Z d d l m Z d d l m Z d d l m Z d d l m Z d d
|
l m Z e e d Z d Z d Z d Z d Z d Z d Z d Z! d Z" d d d d Z# d Z$ d Z% d d Z& d S( s YOLO_v3 Model Defined in Keras.iÿÿÿÿ( t wrapsN( t backend( t Conv2Dt Addt ZeroPadding2Dt UpSampling2Dt Concatenate( t LeakyReLU( t BatchNormalization( t Model( t l2( t composec O sR i t d d 6} | j d d k r. d n d | d <| j | t | | S( s4 Wrapper to set Darknet parameters for Convolution2D.gü©ñÒMb@?t kernel_regularizert stridesi t validt samet padding( i i ( R
|
t gett updateR ( t argst kwargst darknet_conv_kwargs( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt DarknetConv2D s % c O s? i t d 6} | j | t t | | t t d d S( sC Darknet Convolution2D followed by BatchNormalization and LeakyReLU.t use_biast alphag¹?( t FalseR R R R R ( R R t no_bias_kwargs( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt DarknetConv2D_BN_Leaky s c C s t d d f | } t | d d d | } xT t | D]F } t t | d d
|
t | d | } t | | g } q@ W| S( s@ A series of resblocks starting with a downsampling Convolution2Di i i R i ( i i ( i i ( i i ( i i ( i i ( i i ( R R t rangeR R ( t xt num_filterst
|
num_blockst it y( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt resblock_body" s c C ss t d d | } t | d d } t | d d } t | d d } t | d d } t | d
|
d } | S( s, Darknent body having 52 Convolution2D layersi i i@ i i i i i i i i ( i i ( R R" ( R ( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt darknet_body. s c C s t t | d t | d d t | d t | d d t | d | } t t | d d t | d
|
| } | | f S( s: 6 Conv2D_BN_Leaky layers followed by a Conv2D_linear layeri i i ( i i ( i i ( i i ( i i ( i i ( i i ( i i ( R R R ( R R t out_filtersR! ( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt make_last_layers8 s c C s t | t | } t | j d | | d \ } } t t d d t d | } t | | j d j g } t | d | | d \ } } t t d d
|
t d | } t | | j d j g } t | d | | d \ } } t | | | | g S( s' Create YOLO_V3 model CNN body in Keras.i i i i i i i i\ ( i i ( i i ( R R# R% t outputR R R R t layers( t inputst num_anchorst num_classest darknetR t y1t y2t y3( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_bodyF s # c C s/ t | } t j t j | d d d | d g } t j | d d !} t j t j t j d d | d d d d d g d | d d d g } t j t j t j d d | d d d d d g | d d d d g } t j | | g } t j | t j | } t j | d | d | d | | d g } t j
|
| d d d
f }
|
t j | d d d
|
f } t j
|
| d d
|
d
f } t j
|
| d d d
f } |
|
| t j | d d d
t j | }
|
| | t j | d d d
t j | } |
|
| | | f S( s8 Convert final layer features to bounding box parameters.i i i i t stopiÿÿÿÿi .Ni ( t lent Kt reshapet constantt shapet tilet aranget concatenatet castt dtypet sigmoidt exp( t featst anchorsR* t input_shapeR) t anchors_tensort
|
grid_shapet grid_yt grid_xt gridt box_xyt box_wht box_confidencet box_class_probs( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_headZ s$ *44' 00c C sP | d d d d
f } | d d d d
f } t j | t j | } t j | t j | } t j | t j | | } | | d | } | | } | | | } | | 9} | | d } | | d }
|
t j | d d d
f | d d d
f |
|
d d d
f |
|
d d d
f g } | t j | | g 9} | S( s Get corrected boxes.Niÿÿÿÿg @i i i ( R2 R9 R: t roundt minR8 ( RE RF R? t image_shapet box_yxt box_hwt new_shapet offsett scalet box_minst box_maxest boxes( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_correct_boxesw s$
|
|
c C sz t | | | | \ } } } } t | | | | } t j | d d g } | | }
|
t j |
|
d | g }
|
| |
|
f S( s Process Conv layer outputiÿÿÿÿi ( RI RU R2 R3 ( R= R> R* R? RL RE RF RG RH RT t
|
box_scores( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_boxes_and_scores s
|
i g333333ã?g à?c C sB d d d g d d d g d d d g g } t j | d d d !d
|
} g } g }
|
xX t d D]J } t | | | | | | | | \ } } | j | |
|
j | qa Wt j | d d } t j |
|
d d }
|
|
|
| k } t j | d d } g } g } g } xç t | D]Ù } t j | | d d
| f } t j |
|
d d
| f | d d
| f } t j j
|
| | | d | } t j | | } t j | | } t j | d | } | j | | j | | j | qWt j | d d } t j | d d } t j | d d } | | | f S( s= Evaluate YOLO model on given input and return filtered boxes.i i i i i i i i i i t axisR: t int32Nt iou_threshold( R2 R5 R RW t appendR8 R4 t tft boolean_maskt imaget non_max_suppressiont gathert ones_like( t yolo_outputsR> R* RL t max_boxest score_thresholdRZ t anchor_maskR? RT RV t lt _boxest _box_scorest maskt max_boxes_tensort boxes_t scores_t classes_t ct class_boxest class_box_scorest nms_indext classes( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_eval s> *
|
"2 c C s d d d g d d d g d d d g g } t j | d
|
d } t j | d
|
d } | d d d
f | d d d
f d } | d d d
f | d d d
f } | | d d d
| d d d
f <| | d d d
| d d d
f <| j d } g t d D]) } | i d d 6d d 6d d 6| ^ q} g t d D]H } t j | | | d | | d t | | d | f d
|
d ^ qP}
|
t j | d } | d } | } | d d k } x/t | D]!} | | | | f } t j | d } | d } | } t j | | } t j | | } t j | | d } | d | d } | d | d } | d | d } | | | | } t j | d d } xJt
|
| D]<\ } } x-t d D] } | | | k rØt j | | | d f | | d j d } t j | | | d f | | d j d } | | j | } | | | d f j d } | | | d d
f |
|
| | | | | d d
f <d |
|
| | | | | d f <d |
|
| | | | | d | f <PqØqØWq¿WqÞW|
|
S( s Preprocess true boxes to training input format
|
|
Parameters
|
----------
|
true_boxes: array, shape=(m, T, 5)
|
Absolute x_min, y_min, x_max, y_max, class_code reletive to input_shape.
|
input_shape: array-like, hw, multiples of 32
|
anchors: array, shape=(N, 2), wh
|
num_classes: integer
|
|
Returns
|
-------
|
y_true: list of array, shape like yolo_outputs, xywh are reletive value
|
|
i i i i i i i i i R: t float32RY .Niÿÿÿÿi i g @iþÿÿÿg RX ( .i ( .i ( .i ( .i ( .i ( .i ( .i ( t npt arrayR5 R t zerosR1 t expand_dimst maximumt minimumt argmaxt enumeratet floort astypet index( t
|
true_boxesR? R> R* Re t boxes_xyt boxes_wht mRf t grid_shapest y_truet anchor_maxest anchor_minst
|
valid_maskt bt whRS RR t intersect_minst intersect_maxest intersect_wht intersect_areat box_areat anchor_areat iout best_anchort tt nR t jRn ( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt preprocess_true_boxesÈ sP *.*'' <X
|
|
11 6 !c C s> t j | d } | d d d
f } | d d d
f } | d } | | } | | } t j | d } | d d d
f } | d d d
f } | d } | | }
|
| | } t j | |
|
} t j | | } t j | | d } | d
|
| d } | d | d } | d | d } | | | | } | S( sà Return iou tensor
|
|
Parameters
|
----------
|
b1: tensor, shape=(i1,...,iN, 4), xywh
|
b2: tensor, shape=(j, 4), xywh
|
|
Returns
|
-------
|
iou: tensor, shape=(i1,...,iN, j)
|
|
iþÿÿÿ.Ni i g @i g i ( .i ( .i ( .i ( .i ( .i ( .i ( R2 Rx Ry Rz ( t b1t b2t b1_xyt b1_wht
|
b1_wh_halft b1_minst b1_maxest b2_xyt b2_wht
|
b2_wh_halft b2_minst b2_maxesR R R R t b1_areat b2_areaR ( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt box_iou s(
|
|
|
|
|
|
c s | d } | d d d d g d d d g d d d g g } t j t j | d d d !d
|
t j d } g t d D]9 t j t j | d d !t j d ^ q } d } t j | d d xt d D]} d d d
f } d d d
f }
|
t | | | | | \ } } } } t j | | g d d d
f | | d d d
} t j d d d
f t j | } t j | | t j | } t j | | g d d } d d d d
f d d d
f } t
|
j t j d d d d t } t j | d f d } t j j f d | d | g \ } } | j } t j | d } | t j | | } | t j d | d | t j d | | } | t j |
|
| } | t j | t j | t j | 7} që W| t j t j | S( s~ Return yolo_loss tensor
|
|
Parameters
|
----------
|
yolo_outputs: list of tensor, the output of yolo_body
|
y_true: list of array, the output of preprocess_true_boxes
|
anchors: array, shape=(T, 2), wh
|
num_classes: integer
|
ignore_thresh: float, the iou threshold whether to ignore object confidence loss
|
|
Returns
|
-------
|
loss: tensor, shape=(1,)
|
|
i i i i i i i i i i .NiÿÿÿÿRX t sizet dynamic_sizet boolc s t j | d d d
f | d d f } t | | } t j | d d } | j | t j | k t j | } | d | f S( N.i i RX iÿÿÿÿi ( R\ R] R¥ R2 t maxt writeR9 R: ( R t ignore_maskt true_boxR t best_iou( t ignore_threshRf t object_mask_boolt pred_boxR
( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt loop_bodyb s
|
6-c s
|
| k S( N( ( R R ( R ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt <lambda>h s ( R2 R9 R5 R: R RI R8 t logt switcht
|
zeros_likeR\ t TensorArrayt Truet control_flow_opst
|
while_loopt stackRx t squaret sum( R R> R* R® Rb Re R? R t losst object_maskt true_class_probst pred_xyt pred_wht pred_confidencet pred_class_probst xy_deltat wh_deltat box_deltat box_delta_scaleR« R± t _t box_losst confidence_losst
|
class_loss( ( R® Rf R R¯ R° R
s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt yolo_loss7 s<
|
|
*7L
|
#30 6(- 1(' t __doc__t functoolsR t numpyRu t
|
tensorflowR\ t kerasR R2 t keras.layersR R R R R t! keras.layers.advanced_activationsR t keras.layers.normalizationR t keras.modelsR t keras.regularizersR
|
t yolo3.utilsR R R R" R# R% R/ RI RU RW Rs R R¥ RÌ ( ( ( s/ /home/zzg/Opensource/keras-yolo3/yolo3/model.pyt <module> s2 (
|
& E *
|
