I inputs Placeholder*% shape: * dtype0  conv2d/kernelConst* dtype0* valueB"(-| ?QS?ӽvٗ?B?[ NÄPI5?-__1>/> ?D?ؾ*LYdn=׸>'z} ˉCZ8=5? ?%1h=28?(>)UY5"?O< T conv2d/biasConst* dtype0*1 value(B&"S.?[=%>Z"_>芵 @ conv2d/Conv2D/ReadVariableOpIdentity conv2d/kernel* T0  conv2d/Conv2DConv2Dinputsconv2d/Conv2D/ReadVariableOp* dilations * T0* data_formatNHWC* strides * explicit_paddings * use_cudnn_on_gpu(* paddingSAME ? conv2d/BiasAdd/ReadVariableOpIdentity conv2d/bias* T0 g conv2d/BiasAddBiasAdd conv2d/Conv2Dconv2d/BiasAdd/ReadVariableOp* T0* data_formatNHWC , conv2d/ReluReluconv2d/BiasAdd* T0  max_pooling2d/MaxPoolMaxPool conv2d/Relu* ksize * paddingSAME* T0* strides * data_formatNHWC  conv2d_1/kernelConst* dtype0* valueB"Y1+,?M`irﾕ,}#>>7CD>󐇽n>z9>#>CC{$>,]7Ѷ$S=?>/n?ֆ?ؾ+^=W=1{/2k% >,>?+l=I|>=ߦ%cs?@?w>=gO>u?`?+Tc=?F>~>u1>YMxe?zGF:=>>dzPG?̣>(')>[>>;>He>'{{zվ ZнۿTF?{tJ)1S< ?2J>? H>r!o/%">d=#?3[>W,|?'">@5>6(? خ <>>m>tFsKo> >`<>F??0>gV>o=Z!č>.<>r=ui o>v<+(5޾'A>H>ٿ1 "I Ie{E>^f>i>σC伾ڿOξzש,>#1+>?W=0%n V conv2d_1/biasConst* dtype0*1 value(B&"(31aϾޜN>r/>-'> D conv2d_1/Conv2D/ReadVariableOpIdentityconv2d_1/kernel* T0  conv2d_1/Conv2DConv2Dmax_pooling2d/MaxPoolconv2d_1/Conv2D/ReadVariableOp* dilations * T0* data_formatNHWC* strides * explicit_paddings * use_cudnn_on_gpu(* paddingSAME C conv2d_1/BiasAdd/ReadVariableOpIdentity conv2d_1/bias* T0 m conv2d_1/BiasAddBiasAddconv2d_1/Conv2Dconv2d_1/BiasAdd/ReadVariableOp* T0* data_formatNHWC 0 conv2d_1/ReluReluconv2d_1/BiasAdd* T0  max_pooling2d_1/MaxPoolMaxPool conv2d_1/Relu* ksize * paddingSAME* T0* strides * data_formatNHWC  conv2d_2/kernelConst* dtype0* valueB"]i>!$E>q\B? \?-7TUzo>N! ?y>=.;#}>?pR?>%\>ʺb?^. ?D￈?">r|ӽf>?|&y=g>[g{>>O;(>s?V="?sB?b^Lp>YA Q>9->a2W?++g>mgɾ M>R>j?x=苅,庾6>9U>,>Rjggھ2=׫gL-cN>T@'!d.>>87>a#>?4'>g??VinI?]Ăp+?zY =x>|??>\Dg?m>տw_?JaJt>N=S E>.>3;4Ծ?>k<^}M>>ℳ6E>?o>g=> E V conv2d_2/biasConst* dtype0*1 value(B&"Z?7<>E@L)? D conv2d_2/Conv2D/ReadVariableOpIdentityconv2d_2/kernel* T0  conv2d_2/Conv2DConv2Dmax_pooling2d_1/MaxPoolconv2d_2/Conv2D/ReadVariableOp* dilations * T0* data_formatNHWC* strides * explicit_paddings * use_cudnn_on_gpu(* paddingSAME C conv2d_2/BiasAdd/ReadVariableOpIdentity conv2d_2/bias* T0 m conv2d_2/BiasAddBiasAddconv2d_2/Conv2Dconv2d_2/BiasAdd/ReadVariableOp* T0* data_formatNHWC 0 conv2d_2/ReluReluconv2d_2/BiasAdd* T0  max_pooling2d_2/MaxPoolMaxPool conv2d_2/Relu* ksize * paddingSAME* T0* strides * data_formatNHWC  conv2d_3/kernelConst* dtype0* valueB"p<:&:>Lɿ} %I=Z>޾vM>y? =q>4?%*?8?? |z>Vn>6?dI?a.>2Cx7Z5;uN={v?>Ґ??o0">ͨ\ي>U}%YZQ?_=$=`>D>V=L"̑%r=4`?D\d?9B](?| ?w6Vb琻 ?-$>rΊ>>sO>& =D>D=ב>">)>jV=32̽3 ?V([>b _=r$~0I> *Q>yWȿG\; K(8e?hV?[$?Mf"?E?Z xI¿n ?P?<&> ?q>ܾ=0{;)]4;w Z>EX?ll=oS?,?g"w]R >Y>G5F"?7[/?BY>/wq ?yh= V conv2d_3/biasConst* dtype0*1 value(B&"Ț?WK=bu<,? c> ? D conv2d_3/Conv2D/ReadVariableOpIdentityconv2d_3/kernel* T0  conv2d_3/Conv2DConv2Dmax_pooling2d_2/MaxPoolconv2d_3/Conv2D/ReadVariableOp* dilations * T0* data_formatNHWC* strides * explicit_paddings * use_cudnn_on_gpu(* paddingSAME C conv2d_3/BiasAdd/ReadVariableOpIdentity conv2d_3/bias* T0 m conv2d_3/BiasAddBiasAddconv2d_3/Conv2Dconv2d_3/BiasAdd/ReadVariableOp* T0* data_formatNHWC 0 conv2d_3/ReluReluconv2d_3/BiasAdd* T0 d /global_average_pooling2d/Mean/reduction_indicesConst* valueB"* dtype0  global_average_pooling2d/MeanMean conv2d_3/Relu/global_average_pooling2d/Mean/reduction_indices* Tidx0* keep_dims(* T0 u dense/kernelConst* dtype0*Q valueHBF"8Ņx?g?{ܓr1è???{?y ? dense/biasConst* dtype0* valueB" ? > dense/MatMul/ReadVariableOpIdentity dense/kernel* T0  dense/MatMulMatMulglobal_average_pooling2d/Meandense/MatMul/ReadVariableOp* transpose_b(* T0* transpose_a( = dense/BiasAdd/ReadVariableOpIdentity dense/bias* T0 d dense/BiasAddBiasAdd dense/MatMuldense/BiasAdd/ReadVariableOp* T0* data_formatNHWC * outputIdentity dense/BiasAdd* T0