From 5eab06018c51d0ec8bf85919bf50438e2c4056a3 Mon Sep 17 00:00:00 2001 From: andrewghoward Date: Tue, 13 Jun 2017 21:57:48 -0700 Subject: [PATCH] MobileNet V1 commit (#1551) * MobileNet V1 commit * updates to README --- slim/BUILD | 18 + slim/README.md | 26 +- slim/nets/mobilenet_v1.md | 47 ++ slim/nets/mobilenet_v1.png | Bin 0 -> 100916 bytes slim/nets/mobilenet_v1.py | 397 +++++++++++++++++ slim/nets/mobilenet_v1_test.py | 450 ++++++++++++++++++++ slim/nets/nets_factory.py | 5 +- slim/preprocessing/preprocessing_factory.py | 4 +- 8 files changed, 931 insertions(+), 16 deletions(-) create mode 100644 slim/nets/mobilenet_v1.md create mode 100644 slim/nets/mobilenet_v1.png create mode 100644 slim/nets/mobilenet_v1.py create mode 100644 slim/nets/mobilenet_v1_test.py diff --git a/slim/BUILD b/slim/BUILD index 77a1ae50353..348ca759560 100644 --- a/slim/BUILD +++ b/slim/BUILD @@ -132,6 +132,7 @@ py_library( ":cifarnet", ":inception", ":lenet", + ":mobilenet_v1", ":overfeat", ":resnet_v1", ":resnet_v2", @@ -269,6 +270,23 @@ py_library( srcs = ["nets/lenet.py"], ) +py_library( + name = "mobilenet_v1", + srcs = ["nets/mobilenet_v1.py"], + srcs_version = "PY2AND3", +) + +py_test( + name = "mobilenet_v1_test", + size = "large", + srcs = ["nets/mobilenet_v1_test.py"], + shard_count = 3, + srcs_version = "PY2AND3", + deps = [ + ":mobilenet_v1", + ], +) + py_library( name = "overfeat", srcs = ["nets/overfeat.py"], diff --git a/slim/README.md b/slim/README.md index 85275e8d17e..628931f7cbe 100644 --- a/slim/README.md +++ b/slim/README.md @@ -194,21 +194,24 @@ Model | TF-Slim File | Checkpoint | Top-1 Accuracy| Top-5 Accuracy | [Inception V2](http://arxiv.org/abs/1502.03167)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/inception_v2.py)|[inception_v2_2016_08_28.tar.gz](http://download.tensorflow.org/models/inception_v2_2016_08_28.tar.gz)|73.9|91.8| [Inception V3](http://arxiv.org/abs/1512.00567)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/inception_v3.py)|[inception_v3_2016_08_28.tar.gz](http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz)|78.0|93.9| [Inception V4](http://arxiv.org/abs/1602.07261)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/inception_v4.py)|[inception_v4_2016_09_09.tar.gz](http://download.tensorflow.org/models/inception_v4_2016_09_09.tar.gz)|80.2|95.2| -[Inception-ResNet-v2](http://arxiv.org/abs/1602.07261)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/inception_resnet_v2.py)|[inception_resnet_v2.tar.gz](http://download.tensorflow.org/models/inception_resnet_v2_2016_08_30.tar.gz)|80.4|95.3| -[ResNet V1 50](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_50.tar.gz](http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz)|75.2|92.2| -[ResNet V1 101](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_101.tar.gz](http://download.tensorflow.org/models/resnet_v1_101_2016_08_28.tar.gz)|76.4|92.9| -[ResNet V1 152](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_152.tar.gz](http://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz)|76.8|93.2| -[ResNet V2 50](https://arxiv.org/abs/1603.05027)^|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v2.py)|[resnet_v2_50.tar.gz](http://download.tensorflow.org/models/resnet_v2_50_2017_04_14.tar.gz)|75.6|92.8| -[ResNet V2 101](https://arxiv.org/abs/1603.05027)^|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v2.py)|[resnet_v2_101.tar.gz](http://download.tensorflow.org/models/resnet_v2_101_2017_04_14.tar.gz)|77.0|93.7| -[ResNet V2 152](https://arxiv.org/abs/1603.05027)^|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v2.py)|[resnet_v2_152.tar.gz](http://download.tensorflow.org/models/resnet_v2_152_2017_04_14.tar.gz)|77.8|94.1| -[VGG 16](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/vgg.py)|[vgg_16.tar.gz](http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz)|71.5|89.8| -[VGG 19](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/vgg.py)|[vgg_19.tar.gz](http://download.tensorflow.org/models/vgg_19_2016_08_28.tar.gz)|71.1|89.8| - +[Inception-ResNet-v2](http://arxiv.org/abs/1602.07261)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/inception_resnet_v2.py)|[inception_resnet_v2_2016_08_30.tar.gz](http://download.tensorflow.org/models/inception_resnet_v2_2016_08_30.tar.gz)|80.4|95.3| +[ResNet 50](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_50_2016_08_28.tar.gz](http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz)|75.2|92.2| +[ResNet 101](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_101_2016_08_28.tar.gz](http://download.tensorflow.org/models/resnet_v1_101_2016_08_28.tar.gz)|76.4|92.9| +[ResNet 152](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v1.py)|[resnet_v1_152_2016_08_28.tar.gz](http://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz)|76.8|93.2| +[ResNet V2 200](https://arxiv.org/abs/1603.05027)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/resnet_v2.py)|[TBA]()|79.9\*|95.2\*| +[VGG 16](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/vgg.py)|[vgg_16_2016_08_28.tar.gz](http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz)|71.5|89.8| +[VGG 19](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/vgg.py)|[vgg_19_2016_08_28.tar.gz](http://download.tensorflow.org/models/vgg_19_2016_08_28.tar.gz)|71.1|89.8| +[MobileNet_v1_1.0_224](https://arxiv.org/pdf/1704.04861.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/mobilenet_v1.py)|[mobilenet_v1_1.0_224_2017_06_14.tar.gz](http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz)|70.7|89.5| +[MobileNet_v1_0.50_160](https://arxiv.org/pdf/1704.04861.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/mobilenet_v1.py)|[mobilenet_v1_0.50_160_2017_06_14.tar.gz](http://download.tensorflow.org/models/mobilenet_v1_0.50_160_2017_06_14.tar.gz)|59.9|82.5| +[MobileNet_v1_0.25_128](https://arxiv.org/pdf/1704.04861.pdf)|[Code](https://github.com/tensorflow/models/blob/master/slim/nets/mobilenet_v1.py)|[mobilenet_v1_0.25_128_2017_06_14.tar.gz](http://download.tensorflow.org/models/mobilenet_v1_0.25_128_2017_06_14.tar.gz)|41.3|66.2| ^ ResNet V2 models use Inception pre-processing and input image size of 299 (use `--preprocessing_name inception --eval_image_size 299` when using `eval_image_classifier.py`). Performance numbers for ResNet V2 models are -reported on ImageNet valdiation set. +reported on ImageNet valdiation set. + +All 16 MobileNet Models reported in the [MobileNet Paper](https://arxiv.org/abs/1704.04861) can be found [here](https://github.com/tensorflow/models/tree/master/slim/nets/mobilenet_v1.md). +(\*): Results quoted from the [paper](https://arxiv.org/abs/1603.05027). Here is an example of how to download the Inception V3 checkpoint: ```shell @@ -375,4 +378,3 @@ image_preprocessing_fn = preprocessing_factory.get_preprocessing( See [Hardware Specifications](https://github.com/tensorflow/models/tree/master/inception#what-hardware-specification-are-these-hyper-parameters-targeted-for). - diff --git a/slim/nets/mobilenet_v1.md b/slim/nets/mobilenet_v1.md new file mode 100644 index 00000000000..3ce23117681 --- /dev/null +++ b/slim/nets/mobilenet_v1.md @@ -0,0 +1,47 @@ +# MobileNet_v1 + +[MobileNets](https://arxiv.org/abs/1704.04861) are small, low-latency, low-power models parameterized to meet the resource constraints of a variety of use cases. They can be built upon for classification, detection, embeddings and segmentation similar to how other popular large scale models, such as Inception, are used. MobileNets can be run efficiently on mobile devices with [TensorFlow Mobile](https://www.tensorflow.org/mobile/). + +MobileNets trade off between latency, size and accuracy while comparing favorably with popular models from the literature. + +![alt text](https://github.com/tensorflow/models/tree/master/slim/nets/mobilenet_v1.png, "MobileNet Graph") + +# Pre-trained Models + +Choose the right MobileNet model to fit your latency and size budget. The size of the network in memory and on disk is proportional to the number of parameters. The latency and power usage of the network scales with the number of Multiply-Accumulates (MACs) which measures the number of fused Multiplication and Addition operations. These MobileNet models have been trained on the +[ILSVRC-2012-CLS](http://www.image-net.org/challenges/LSVRC/2012/) +image classification dataset. Accuracies were computed by evaluating using a single image crop. + +Model Checkpoint | Million MACs | Million Parameters | Top-1 Accuracy| Top-5 Accuracy | +:----:|:------------:|:----------:|:-------:|:-------:| +[MobileNet_v1_1.0_224](http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz)|569|4.24|70.7|89.5| +[MobileNet_v1_1.0_192](http://download.tensorflow.org/models/mobilenet_v1_1.0_192_2017_06_14.tar.gz)|418|4.24|69.3|88.9| +[MobileNet_v1_1.0_160](http://download.tensorflow.org/models/mobilenet_v1_1.0_160_2017_06_14.tar.gz)|291|4.24|67.2|87.5| +[MobileNet_v1_1.0_128](http://download.tensorflow.org/models/mobilenet_v1_1.0_128_2017_06_14.tar.gz)|186|4.24|64.1|85.3| +[MobileNet_v1_0.75_224](http://download.tensorflow.org/models/mobilenet_v1_0.75_224_2017_06_14.tar.gz)|317|2.59|68.4|88.2| +[MobileNet_v1_0.75_192](http://download.tensorflow.org/models/mobilenet_v1_0.75_192_2017_06_14.tar.gz)|233|2.59|67.4|87.3| +[MobileNet_v1_0.75_160](http://download.tensorflow.org/models/mobilenet_v1_0.75_160_2017_06_14.tar.gz)|162|2.59|65.2|86.1| +[MobileNet_v1_0.75_128](http://download.tensorflow.org/models/mobilenet_v1_0.75_128_2017_06_14.tar.gz)|104|2.59|61.8|83.6| +[MobileNet_v1_0.50_224](http://download.tensorflow.org/models/mobilenet_v1_0.50_224_2017_06_14.tar.gz)|150|1.34|64.0|85.4| +[MobileNet_v1_0.50_192](http://download.tensorflow.org/models/mobilenet_v1_0.50_192_2017_06_14.tar.gz)|110|1.34|62.1|84.0| +[MobileNet_v1_0.50_160](http://download.tensorflow.org/models/mobilenet_v1_0.50_160_2017_06_14.tar.gz)|77|1.34|59.9|82.5| +[MobileNet_v1_0.50_128](http://download.tensorflow.org/models/mobilenet_v1_0.50_128_2017_06_14.tar.gz)|49|1.34|56.2|79.6| +[MobileNet_v1_0.25_224](http://download.tensorflow.org/models/mobilenet_v1_0.25_224_2017_06_14.tar.gz)|41|0.47|50.6|75.0| +[MobileNet_v1_0.25_192](http://download.tensorflow.org/models/mobilenet_v1_0.25_192_2017_06_14.tar.gz)|34|0.47|49.0|73.6| +[MobileNet_v1_0.25_160](http://download.tensorflow.org/models/mobilenet_v1_0.25_160_2017_06_14.tar.gz)|21|0.47|46.0|70.7| +[MobileNet_v1_0.25_128](http://download.tensorflow.org/models/mobilenet_v1_0.25_128_2017_06_14.tar.gz)|14|0.47|41.3|66.2| + + +Here is an example of how to download the MobileNet_v1_1.0_224 checkpoint: + +```shell +$ CHECKPOINT_DIR=/tmp/checkpoints +$ mkdir ${CHECKPOINT_DIR} +$ wget http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz +$ tar -xvf mobilenet_v1_1.0_224_2017_06_14.tar.gz +$ mv mobilenet_v1_1.0_224.ckpt.* ${CHECKPOINT_DIR} +$ rm mobilenet_v1_1.0_224_2017_06_14.tar.gz +``` +More information on integrating MobileNets into your project can be found at the [TF-Slim Image Classification Library](https://github.com/tensorflow/models/blob/master/slim/README.md). + +To get started running models on-device go to [TensorFlow Mobile](https://www.tensorflow.org/mobile/). diff --git a/slim/nets/mobilenet_v1.png b/slim/nets/mobilenet_v1.png new file mode 100644 index 0000000000000000000000000000000000000000..a458345174a12073a653e26d6747914a4e58e516 GIT binary patch literal 100916 zcmZU)1ymeSw={~o2KT|;3GTt&CAdRycXxMp2$taPE`vizkl+r%CAhrK{r~&!{oZ3O zR`;6eY1zHc*|n?cM5`!CqaqO^K|nyD%F0NnK|nx(fqzH@IN%cv1zF$=5QMCRn1+|( zX^)kg{-OK%`&4$s1|o*>MNcdReKdncraqP4(`16a%dSgmHC4h5`(|oOqt;?4&q(w| z%r3=BBb8o^X0_>e)eBp~IIM z;n}V>M$*Kw;#7trq)Mo7LI1Off?|O+KSTWA9{~sS2SFi$e-mjyYKr2h!2b6pmqNcjIgpUTpc4GvCZAxo!4a#;ft*;tjoH?{UOB43ps9VaeY0;{hBNK!E`R|jbW!7 zB0M_j@JKA)_LUb*KqM5$yP>*zsA4X+$xI&K<@0N{qy2iw|Gb}nFCz?2PG#k?#>Y6q1Q6BQFMz0MCg6J8RXUyHH z_uDsyYQrAFCaYP}TmOSxgSP*?2wMt}HJ{Q<+%h2)&^gXiqT@y|#kk&4pQPA+m4SIr39eOG(XzL^S_b zre;a}iqax}-S390KHQ@bz`2IphEaZcIzv0W; z`@DW(IuHU2)fcA!ltEw;aJOPbjSNRQ8HelY4aE)r%Dym?LM>l!yTl@UBW0~LleX=Y z>BjvZqd6Fbt|;hKMPtz3*7o}{ZPs9X=tF_m4&oI)C(=}O&=dj(r?KD5)!2?0jUbVL zq_1z6d7rTdvSpu6%NHEdz!zs(bZve8myQrBG*3k<_5XL0`S&7aN`dr18!_+QMaE;1 zd1^%$eh~f9gqh%Zx=fB~GRQ7+3&Q%Eo144ui6wT>*VWniH45Q|vhVHb#M;(&h%!+y zgf-wQR;Sf2yHvw*JQM*nO+XMPU-5rUWssWv!T+uyL6PR-wDI=(eC7Cl=a@p9J-=HQ z@b`PlP*&(g3a0@^3W4*%=0iUHhrzyu5a;tXzAL(ml_vPfOx7{8M4<>hhYr}XvNE&t z={XH&GnS(Nn6KDSalm^I%&_M0BoVEMzo|F8c2Qqm4n;B*umc}61L+Bv@P(8F$R>&}0qQU8PA-mHQm z@VsdjCK#4qz0NxhsbUoLMtFL!Q0>;*=Ldci7KSgz0J)*uJ=N^M_F>x1>Ep+bX0@{t zNudL5Ap=o)GDu*viE>VJ9moY%e_@jViB&?!;|-yN z{{Bob7Oda=-4qbrBTyKOA+nd0+E!C~tM2bGFL^P&uAbK~pL;I7)(C6~fbjhIq5Lu? z@}^oc7PrQ6!OnQ7siASZN;~x>MQD)Hz>Ofj96mdnMMr;FBI#J$gTtAqr-%FzL;QB< zT;gLsZ7Nb$Kw<6)4#e>vB#IBw41VTnZU${;unO`ar=3mk?4?~wu#&s6xK#AncXE=wmg=4X-y>$C!#{_R(ea`~Xq^sNh zy&nj&@G;BZvo_2MXYV^&Zl1Adl@N#nZ|9{smJ@@z{hvM3($aXpSQjY7&A|eBl>oDl z%~2Z!i2>wlcRwS@O5bBW=o4%TB%jD8++G1Gta;@1u^oX{KLiIf3XJ)kBJAEgWcFvw zq>mwa+cSK>|JXztPjEwBuQXYWcq;h9#`-%E686;8)xism!dR6%luECNHeYdYYOcw% z_E1&rWF4GsN0Mc9gGXlt7q=NeiDgLvh4#Qi-c-}>z={LxsH@5!PPYQLHiN@sriY!B zZ<~?Kw}OGkYB+oma4$fhF@|ta__|UkJBm1X;S{8hcun?p2_MbxN^02wo}JB%f?eoi zpC(!u!OS4|7yvwYgTp#M!NcKChT@>S`WL|jCZu!sI&=oI8^kium6xUkqP zd(S8lw_QIlTT{QYi{+fS^3#I^H02H_Gk}w>sC{nZBn6&otTcxul^NK{XRHZ2)&Cwh zBimk6Q!}zX6e)HB^Ewu0uFmgM`JN>(M0fS;ySDTG1P(pSV2n46Jrsw~I4mse2Tj9v zCzC)-%-S)V+pJX;(HZNelZmlnprZ3iEJ!w%2SK?keqBV3{{#tZepmwB!xBTOYa=;t zB?|_E3>7aet@P;!ILpY;&og7nbl6Y&w^^Vm8TL|en8ktUK^hV<@a~((-S>AjL?ONN z29Hg3RYf8B7&%U)Mgh#y=(_#yO?;$0{E2WKXLl#-KZgWP?ew;MnV|>GKCrM=!2&b* zh9D5emYif)2GA_nvhXK-O^{$!uGQ#89@KT*1yt(iAUmi=KD<#%nQx9=V@>f58LFbB z^2G#iJy6#u7Z9nJmpDnZCZ$I#e;JE{s`E`a7?bElNl_!t?A37IHs@TXqd%n9Px2?i zjPgKb0f)LO7r;E12MHn)ig}tO0pb2lPSXTzQajw;Hh-XkKMgEbZ@y41e@K{CzVYm$ zMi{xf4>JoGYh1dH7j60eZsjs*ft-Pp^w*C-6NuUwC{+>2i19^=PLXk-GTR1bO(B^> zC8d&QQ!hT6d|gf;9HWkg;Is=Lss$XpwSa3k9pT-#rLQGk{mQpi=!w!(Z z*(&aYHR_PwgdY#M_^@I3Y)yvq=NSxfq)oG=i47JD4 zPM~50CC(iy4y}?0avz9-H_s7%U_A+paq)Qyd*UO37g{VT&YkH&%ImGUh`zN!f>@uu z|B&uVK=ARY<*m)0MX?9-B-GZTadB~#l$9YG7#OG(28%o%A0H<-H%C-~)fA+2Q+K7s z#ELZ@shMtuN-WdUn^aZVk%bG%5K&SKS7PWGLFfy!)2W!>f4IB3$)xjhvjke1T_aXB zHmU{CXm?EBE=3oIQYqqM4Ow4e35 zOvK)vdwK2v00#MtmW%u*Y^<&O=S!vchrYNo?F~eHuc{hQYjS@IviLDIp)c^b{ez z@%JCN+>U=xcVAC1K3oo=&#$aRcXtb3P2B$dJI)B*3)Ozn`+Pa~3Ec#MIUi(6Yl&~Y zh%6hKDuM)cKa1ZF0obM59~cQhBap2peQ!?@WXGY}zK*+d#rEgq#Cd1fIgkHumD2}C zNUXTX&!D`J6DM3Whg2~cj^)#pCTg}0(-AprY;1YBIP2RR!y~trMyP8xIS)_IUC+K} zk`4cR1IX=oA|BJLoe`-1+qpaePo>`7)an0%CIBY<;dP95QSm*QwviNeFS@c~(o!54 zA_|q?P82eMk#?-6wv>tpI|>b^Aw2cy+Pcy?IzJygUZA}?uBaH6o8Q>rw&UXABRsv; zNg>akM9EXe^%E}CVZ~2dt2ez{q z8TfwjJ_5Wq=T-Z*@knKsJbZ&%VA8yq6-k7KM>9U0$$z*Jd3R+`V+08Z!MX`+8ovPJ zHql7Hfa3<(u`)DUAAG=(!AXjb4^}msj45qtAs9;}ORZ!lh<5xHDHXRcF8=9t+)vL) zMI{JGEVH2qRGn@wdKXvZ4@q{1N=i$2R#YHPalb!5>924de>Woncq$Xp)AEbG{fNlh zu1E~}BT&w8MJ=iL9-k`;0|@K--uNAQA1z-lBSq4%sgJ*;U3tFVvQ}a4-P(S$&C3E} zhP9xaIa3YnWdG7$5@v^0gEa4A9ixsqav7oG9BD9<4$IH#4uTvq2TfB!6Q~?2pQCSX z-AuoY#^*NJt+?zR;sM}6mcQLi>>3DILiNj0Y2xKhEUPKdmbLF^=+*oD{erlDxVHxe zPX<1>1rF^`W;S@=SlVs&8c=}ku*Et3d1i&5nUo5J1n0W-jG+cp?lSZRL_iL2K`AQe^-26GI z_H-ZtRoTPwav+2&8hUN%-+-MehVmOLbx7JE!7V;Za&)}kzdt+O9$^v-dWkmrX1lLd z78UW0AsmO9yR0tOtk&DD$ax70(@U6AOk>S^NI`KB6>iDTt+zRnp^*tQ6_^wDA&%QF zRwt)9_}-lkZ1nh2n@)`bpw5JbYtwznz_ia8JBNV9LBlFr1|Hv57j~}!`^QC74N!wy zahA@?N`M`PFUQ?0TE+p$ZFLE)@jIpA;4?u#;$QC#Tk4)kF(A>_u=A^dY?&ntx2Kk0nERS5mOx zg@&-_@!b&IoxjR#w5A5f^?7o@z@Sb#Z)Ve(?2an_%O6z@jRHDFp@&Mf^+N(g@OX+U zgj8bow=iA$hqPfPT=`5E9F=m!JY`xxx$Q_*^Uut@m9u+N1v3>n|7QyIwnGA|H?3Pv z7&3xp(=rNj#`w5FtD$S%hRE*jdnL+Lwvu{!t6xLRN`(_6?M6kZYwB0qF8aS|+9qdO zS+yFgVsX-g5}Rh#j0tB&UT@VKG%szUdrhTM+U!?pit_TtP4jZIU1EU=yn`n9PM|tS z@GRF$j;sjkI?bWirthW{vhkIqKiIoEr7iGX*zbDW@aL~zn@*co)~IHJ3w|ekrXxse z&SM5Cl-V80x-!EGRS8(R4VdPxn59lq`n&AHfp3C9Q;$VZGk$da@_0j?3k!eUFvayc zny2PZm>pwP0ClZl*K*u53)I&x|B9!1;8yx5F|SKNI^AT)nLTwa(9-M zNp^!@1WJc3EG#IFfJXXq0G=$0Cp*xJD?9W7mqnir;A%)L_@SDhT)34S*QFSMv071* zF_ZhhfOG`(o)d3MkPLEl2E4@OhVRAJTyf~2^c1d%mlcPLQX^jW?(oCt``zkAe+Kz;8h!l#toRQ;gfJ;;d*O zv1nleVH^!I)^$^5s9U}?B8M*)muJ~I9?rdU&iW>*yVBd!PqkG@kg1wY1bA9c6_pVxwjtBPMdeN zV$dbx_#?+3L-{g4ebT9DYHHf_6kK_v zAxJvTv?S;8Nk$wWC^87)=lgIjQ3njtX07h$^R9;x)&O!?k>0zOI$1-r80o$%*I6Oj z{7}tn(trmO%p7s{>nXG$2ES&DJV9@@{wJc%hX}6}T3>jVALJ1;01_ho<(91gaE^Jh z?zH!vimSP~f=0QcyB->JS8xvzWI(Th>Ik(Stcs1#9HJn4eq=q^H2X%_(xOe}BbdIC zkQ;_9D5WK18FT|AtT-F^B#c>(L3IzfzgwL5ZS=r?b!8!HsDS6y&5U5rJRPa4Y^#D~ zjs9z66N@7bJ}kN zhMwGuk7CB==8_4eIJalrb8T02V@4JQ`B7>t+4E}}^NLzt&L-53I=?He%qDONYTKruNS<#qpuVQ7-I%eBiW6dF3pM5%$ycc7DOXGLyz2mzd;6zw;mleJ${fIb4GpAZ{>|_J_RG9=P$x(tcq|a= zlH*9Q-{?j)tP|P(Q{{zK?XN85pdi&n%o# zIr{)Tu2TAs$HuaM#XfCj#sjWobj`5jXU$l6mOx0YHGJ*8!Pu_?M(d-Vgc%jQk}K=L z6+PFcdmU12-&oaDqtK2=SJiAzqQ_dO_7Vq33!OBh zHtH_C&~SB}q-pk)LvFj}sCfUt(A(`N@HHi*iAtx~K`Ulo4&pqkP+iiCtaZX0 zY^d*V$o8`5_74tpU2xdGEA7_{k_?Y+Xm?_t=xN$Yc*LNjk{{7soy;4Th2%^9^QV*c zTR%96cqpAQ^Ae!LFk4zOF_FU6LdK)dwfQr74nb!P05HicArA=bHSbmh3cwFeKLaE% zXE(IM;E=XxL0CM;j6vgSiWUMI9Mqa2vY0k&EnMxE+iC^WSkJcMut79~&X7cqd4Ivn zY^(KLto{~#|6obDj#f9yuQ$HS5myW})MXwHf(I2*QY)*uk;amKFzFsIi%~C#ptn}P zDjnOgs^478V`E$D6zr%!nz#b}_dIx3JF8yj@^cBPd8FDOyHa3QseUzU1U7;zpUwTA z%^XbcY-f4bnl-&B+HtE#<_{<<&%R0WI&6Mrj-;HO<|nY3x!*H$zA~+**9|3n*N-TJ zUt#8e9p?ck0?K2h&y?z8=jN;Yv+MxD1a02LH_;hSfMYs$Kuhkbq5*0eZsrw0aKx;O!*;knUltaP@cby zy?l?(g^ExtxetLcM%&o`*eA`K0TXQnrv;{ypmC-@jaifwX#~zsil6Gn|Jn~}g5U@u z)i0eAvT+h;=-2nm%NYU3U7$)X42A;p+|%C^=fnK2t(*Pt@ZKr+)AVh=0A{NlFMma2 z1Su!daL*f#JDz%w@0<_K-wpBp^g{@C676p%JNSAR$tlWBpQ0#2KTAIragC~@*N9nL zPc$F(`sDr%Aa`o_#2Lna4M?d~Z64=)z1op}NDr8}5=noK z7gdblFUXZzjx2mPDsJomeJ6HtQVY#6Ui~nW0I_&xJX01eQewe&_Kx5R3KWa#P^AT& zL^yCwjpA}&AS)WFAl*1}IS1(G#WH8(Y}M0vy))ja%RgGIC+EOE$K&jjECb5r?&A`3 zLM>1rZ;#0Bq@{fP4AI5U4T)_;!cy&N=P*x+vA0QBo3sBe^i9!S|88{GF=~XYO~Tub zGO%I9QZHx5g@#qJ#&RAw@ffo2z&J)KzW74>RO=2}UXHakib~fs-hbLEOpjdSlLTk$ z!6-+Q&Eu=*OJQ348VHv7N#JV@4x8zmxuR$L0BTZ^0nNpy(k5=7wQWlI`aY+@nybqu z!fu^++TtC|7|=oA*!iYIT|f9JWn}Kq%|AW){8L-BYxOZ_XXj#Dy>YD72-#oP@;x`` zm8BG9$HpU;A^u53K+lMxsHoV;g3DV~T|K-r61$%Y{9T`XfZb#AL|R|_r~W0<&wni( zkVxz(6)uGD(Sr9d3PiX1Jqsd?y)&wyA>5}{$f3Lb_9J~+*BxGPn>cI`dR4|VjVU!~ z%~9T2@6vl+6cn=2RCClZ7t`Ou0z=e^VRtUs&Kadc#~@}oc@bOzh7a^+abZAC5@|%>>~E(esXYN$zay| zr@2TjTUzYH>D$yf`3B$VAaB$J!s7|B4o-jYiMA=J zw=4ncAO6@WTh?^Nb15B(l7NFdDIE7M7kbQ>~DBC{+{0)&D#+;Hfb*P&9FPl_W29E541mYQyO6;(fW{D%5%D&Up?)(u3E7Cw-@04 zxxs#Q+UsGPl^oYqK89yk(F5s~J5-F6=K9+2S~R(897j1IGqtKF&a4MJ9%AB;Pu!g4 z_hv3iaNN_63p1&B8tEi@Jb@!XR|YtuLL^bg)mGI1Qkj0ewHK0*kns9Fux~fjZaf4&Djmwz&r+E z%LE^ZzBJ4CcoU6!BY-a@=J9LC;FmVX(hC@kHM`xEx_kBN>x--*9SZ#gWtT z<}9j_EZX$}ZmHd>e%d-eI*^PQHyyAk1PbM(*({YQGu(^reA)mKMT?~TabQ>%etz!J z{jh_qrZ}6H>oI{mb5FtPing4nuZH72mT=a7GvkYabfuWmGrb2{V|Eq}#Q{QWK&(PV zMKzVo;3Kgbp=dDz4`^Uc?yJL9wjL= zsfvI^(NxviWRD7Kdehb6kz=cOtz5gZ+knwBScz;fFs4XWT-)WK0<*HRC|$^in>c{V zbFeo%P@~yqno&ns8%1lvh~I8gZlqBVHKr zWa9@k&^pc5m>EpE6j~AC$ap2)-TKBUgEri}TBC%I5J%ikv9lm%O(C&2h#CG3id}!aM z)IaOUX+XH{9`tT6`=h1d^9?g`8QO!!7iMK~sg~j>9@^jP^ZezQtF_lN6qL3AtPAz4 z0Y%$tR#4v0yp(-t=Lf6((m!{wmEPt!-N^D7H=jKOM&Q-_%1#P+ynW86)>D%yz#7Ar zd31z?e=bLf>SVFAAtj#U#VxNRD!x;vKxRA60b+Hq{i+9uwT{lYc2!SqpBFm`E9fP6 z^97Ez_n{t3P&+~Tw z{&*T;);E6k`a)2IQJ=#5^9vxWb~ zQ-h89-d4R*TE3)9SUtLlBPf!pWcHrh)R^DEd+cCYd|Dl zNq0U)qHT8^7C;BOZ1%pMIZg%;qPJY_&PBeUzh9z1JnSZ30iwaU-z}R!HXzMae!VeQ z2f9zYZ$kW_Cg!hm@E5h`yo0qE6tx&7R|p~V!Z0O}(NaG84tqY=TTOnK?cJBd{!U|p zPMCSjMbI@~73cPO7rY{s)wdD@anEBXAphw}`Hbz_9D_?~q`h=oms^4K${#v^Dke=$ z9G-4Td`nokdc4t=-VW8>0)9t%4M%80gU!ws>Nja2vT?B|&Bz_a5OD)NIb+(UKLN@z zG{JP5lK*H+U8l-#Y<#pNJ{iZN8Up5oZE*6ZhrLw)%9eLeT<;4M@ig&^IMa$rqR* z9|_Heb>bxoW20H(kC%aLOzOgt=|oJ}2KXd#t>SbcLii>E_FSFvk4c*Jg! zkGV5p?+ z8!l=Osx@jW1%}C<7w9D3hVwHCQa3E!Ed9`FH{sM|ZsdyhuOP+*U{$n z#@Ls#P{$ws;BhUg-mA@e6UtV{K3E)_bcApU(>k_{1r4&WR7&vU<6Cz2!8DgnO=S8!uOrRcYyz;agSU81 zqEX|*1MW8hPpOYcd$C3@6us}S7qQAdN8yE9@&9sm*tV{R1{b9^_IRhg&`j}Xm=i#c zXQ(~-w_>wF46dc|@fo_g!@_z^jR@&IvtzYU@@NYe`9fp@BCxdp%f0?!b^>SRMU^xa zDP|ENVU28pMvmHC$GQGEQ$wR0&bOOym{E6IGLx!#4g%{&OcJeQrzpGh9~FHi1tXx8 zAv5q;Gu0L(Yt~~XdVjqVDn&S+?u0F(&>`4r&}RVe2&VkQS8A)rx54l z$h-JgaJ@p|wU=D>Z_(e3y*q*L_knn^@ak>s&oE+qe7U1~jy>UpZ?_fgevge`9XW&S zT5XkJq>@lUj?4AMWJ!GWj@8j8M<>XW(l>z9B_p5^6?RJU_r?9h7PW9UJ-|#q24h<~ zkj)X=k6N8oIJLf_e17^2zaARV6-@J~aYawclhK8tzSd6Y0B`1#y(cpo51Ev$(f9tE z*Pp3igFg@{)MyMCvvCA>TY|s&xHattzwLaC3fI4$_+{B6@h6f?J`<(AyDA6t$%RzQ zD#&htB)<6Gjg#f|mhJPMkz_M1(_M_)P|qnRGQC7;Wv3(kfuB1n5|0M?xhG}LB}IjA zu1$u4_Mr;o5G)@bsYzP#N3_#ZYJFXOC=BhoMbpjR9#g)v)6>(T8W<7_H+00lrvhy839Z$L`wkGOLTZR3Bt2+Bz>IxVTk=T(y`LE)Z7R_%z*|A_s z0s9LSXARnIJ4V=}`A8IN;oypBnC54K2ZL`|OJcKZXIRbJ82#@%tZ(9QO9U4U9V%yl zcTwd1ahAGd8Z)<9hb=_tXQ+ZLxbN+xPmUe)Gh^@LwBruI^8R7f6j~g~dB%>XNtseE zYAX8{K2wK;QW4}mJ)2V8EGrL)TJ_uEB{2?s{qJAS+n z@sb!x1mxw-b|bilagLPA0bQkyVxF1nJeO)_ zy;%4Cli_EkP()6cQ7L!m!Iyk)Y^CzFCgVr^C6!#w-IOcJA1eUo`hB)g^zhhHA{4Vh z`zw(aJOp3F55-?uH%AN_8!h;Jny|8}LsPem2DTnJ`1m5mR=EuqXN7HZ^~!7@JS5m~ z_J5*C9j)Bj51Yu(np2Qk$7LysX@q=rCUzuGZ&bbmNgpa_HUuHsqX{3Gmyc`1t{OodCWCbUra}#$oRsnkym)UsF(JzVOVryCa3Y{fPy%P^sG*d2g^8*Caj2RalPD z5q;;6qjJexd=tXYR(W6?JSW2U3nzGSMRbQ#S9*KFT8gxz^gep%ItaDV{Gg;+YJ?;9 z7jA>tw7&Q7HC&8A*t*_F2Dv6FBaBz!uFF&)MM#2ei9K8X3GHj`uuYblgK}x#l*has z)X1+DsW)}x_GeKPQ*o2%xr?(r;5b2v{gah!ul;`f+d+a*tRhq912W9aNOmw%xdeYL zBn2#EisFNzw-Jh~l!%k=&$})rwgjZU75ZJv5(JH*5Pl-c>qzdH7V^E5!Nfq@*~LmD z@Ee<%PEc5W2l^A*<7FluMRX2$JpSxj$?Gwap!7{+=6)d zxi%PVbKNo*)Jl@pYSPKjh9n<@Yi2TF3Q&^9BZ5(Ktrb(~-13WzBAYr)YJWjpqVEmg zd(d0@1S85lSwSkBsCceeR!20<&m^%!M(K$w=uD}0xJ<-@9K$PK^-jMX-FlqzWaVQ=OVgcWG=H$e~I?gb@_4=)1&jo zu$r_`e5~W3o3;1$r{~0c!elJtiYhWumHXJ?htkv2Wq)%?#oAPC_2~{I7~7RJ1BEzZ z)BBL*1@Km9FyT${i{(l(V2R}h?ELdZc)dLyj<~+?1m5s`ASv%LZ6EN2ahaL*|K}_G zFBdl4pX}&lN2%6qw1mlp$*CA$7zf(at69{)^nAKnhpMm~#s9pO4Up!4K~~~6zWFqT z{Ep@9$_Q4S%DdJlm|Uvq!D(#{mZ^Jh~H<(rD-V>-lK6S-qb+GH( z4q+w5Z>RW{=AHBsJD$&~r;Jj%4g|P+tix;F9dz|E+cBMGN8#$|mlEj;cOch2fk4_Y z-t}JHf*`w=zuVX9((}#F6i>DHgX#Fqyz~CV4&G)>mzsMfchK^%N&WzFfph|~tZ?LC zx@ZHj;`L>PN1D)x%XqA)r=QxQlGUg;YA6z2mMz(FiqC?wE8PC7aXae;kcFcG1TH_b zJ2An4Tbp@t>OTVmngv3xV01lc^!9W@{1d>t^L6?<{6WiGP-bk zZIidYaO85pk#`_Ua6Ya~GnRn2bFR^7uGn9)jhIWBaLPXn4 zXtDq<^-|KoAlw~%b7Te4(NrHX9 zMnC4*j6b|y1fu)ID~^aj3E5MI-KEkhBW(r#{ce;VLfIV2J6&HKbo%wLTY=rbMJMbn z@c{=?e8vLxkxMH>&RFk-m90{hw-0Ma4g8lrXBW@voz$zOqg~ps&seblW!=ze_tE3m z8nov!ZTj6{EvU29;3(RI0^|4-+bFsbvTFcxz+@t^Cz^|32VHZF$`%WhDU%E4r?lhA zMtv4Oe*l~OK`g#%X{z+y=jxnoBPUIe%V|?kfP$XKN7B|XzQfZh+BuhW(#8Rm1mz)i zI4<+6!|s{OWSj&=O3qVQTUU-f-W)m|OzG-(s4*rfDA(%-x`?vkh&Ww`K11Z`4N6E# zj_7o{MZ=H**21`u5L2gzXnMC4C0hw z{+Cbe6V8C%-(c<}or~)j3tMXhuFR!ffC(v(6FP z4ijk`O1m1l{+4Q_g%(J(^B#9O{nks=Yc0H^DsTb~U+%55F^L1qz|g8B8|ezCg|cO_VZQntAq*)`Vq z$$ll)ieD}Ask!hIwthz|hvLJUSI(TShT4TS_pLjHtPC@MO6lt4SJCs86@9$k|IEe1 zgXozhY?nZl-ka)Ilj6GRf}FaN9*1W#o>Ep>TL7IH>78&a*$V$5!Aj4Jz6NtIY%IY8 z_6A16H{|LCjSwZ}vJZowkF~TuOm$aUsSFFn9^aub^c)}6(n8R(=HWWCkBPcA656>* zkSJ-gwYgb_pI>SKalF~@65raiLNlxt$tP7k*z5+T3^%qxiSkuA_n)_<{B923*RTR% zR6!5BZ3ki42P5h)lGms`E7Tebx{HA~xaE2~!I!F$xMi>%w>#wi&!(h&QJF}W#1~)A z>ophkc<9(zJhQ{oRe&jy78FP4GP=2zvUx_L*37jCZ3=Rlx}*7!w141u7(NL`T^MgD z$4N4CmZ-bm8caklvYDlp=oBY_HlJt19a{s4ql1QDm#nPt(>M*4B;keN*Xd{neq|4A zd~^^IuFL`2Gyhn?=G5bDTWqa6A=?^rN?{ z${w?L!D?-%?y4_A*NJB-^YUWhGjB@|fw#lT7;k;O6S+gSKheZm(^gFT2k+XmHh1Wy z!Qkr#O4JG2`{^IY&gIBSu%h}b54-&%@177#Mk1l2pW2St3Tz*(pB+J?`G-S;j`qW+ ztaU+}*^w%R`2&NZvFfO9z2o2ey&%1LWq>6vvv8-<_TQj z!TFu4t~=`(;+1G2Fc(YdAN59Mh;s4|lB8(Ghi)nmSff*dZ*P5&2TTDv5?4;Lv%J^x zv^EfZI`|+?GAR?Ab)crnvqXXdOl)p7R1dkgohU3+x|q)>0CyyhJry2p-eA0n2^x^K1l5_Py5Q@)fBq+o(zp{>ntAbEy2F_Cli2+uZ<6tVA zGW`UffE(vHMF5tlwYFC2+ZuZxTWknh3tPM;vwF8NpmwnP~ErZXqm^ssPcWR!8zWE$2f2Q{K?}w?3yfFRe7y|Pqnh_ z#`5114J(HG3gyy@W$3VVv7dK_gX4=>m1nh*X_Q+Z7E}-jv2|gn;(t%1q%W&9KwcR5 zqW|sr8#R}}4pW_fwzAa$x6Z_jzL-`skl*A_!p z_t0~+=83ap_<}u6W+c|m$;;a)Aoci1?Ag9=y|t$}I%V_Rk5#Ts!)(5dhuqVhx`Z^- zypB!vxvHcm^)CDP<||QBs)@@{-%23w;y`b2{VZzjgmod~=%&MSjJHtzq|vH+qbJ|0 zeFdhuXruPak+eukjNAQYU?28c(YpneXn9@4@$!QSOE1~W7W@vLmgtPf{jalW37{U1 zIs6T-MBB{c6vP+vuSA2-McYzaDpTF|I#Qh36D$9_T|SZP)Ve+pflPm$QpDkq!_Yy?7p~Ta|bN2Oe6^573-=8gI z+QACYjk@naux7lBNyo}?VfHHA=nmsfg7dah5PDZ<*agi6U?95Z6w) z$cFX2MHeG@MrnQ{Q)pNJmJCsXv5vz#s~M6)01J*;tObN_qo$Yskw?p}%8eNP=2VQ^ zcye~}J1G)a$}z*yP`y{G8mD4aS)N)*Dw#F~nGdk~`5J08BM?n;nlb?3GyXja3Um*9D7j*kNsJVbBjs9wtzT^9+}SkxeqB>l&hrv#k=L@NmH1AK%q zUfBa78HieQM2Jaza;c=|Xs2B#wOyA0hBSiItrPerngKAiqt4L)A?m$6MjQ%rP|l5isaV_$L_x$^Nv{JE702$=_qfJ4)zix=otWoQ z=<`uRAS;Tp>M<>bSJ$s(NmbCUVqIJDkPV9v{E0jzS7$d+liRf_fmGfX zYy@diNUo#YgBG~n4ssoL9uEtpm6a1U^31ZWXv&rD3>(Zo^6MPM5d!RiJ=@gUl35GN(OaQ^l>ParBDnen>#XAyz zOO|l2FT(@MEvoAdp`mi^r@&H{*g@dAE!x$A?o|Fr?o@+SycmSC6?+%))vg&O$W$s= zKQXAGj|-3QS2&$MAF|}h_pL(psH`z@D#+2$i3CQn#5QDwKl^}8#*+xeandhQ%Aw+T zvC-qT9gYVRf;hr7KNUh=eKeRiHf1R03t%ZG*;g=S;?~`&I5*7?-Va%5yI`h^ZVz!irmufSBqGdp>uM#eq0U*5VaD z?Q~a-@l*f?BDaMyj*5Oa_`TwKTQ#bI7)pnE^eOFqk8F$R_xJrdR%gy8W`j zbk>;8-p{ADziD-8yIJbK+U>f){5IyIQ{8vsilcjvT^)Zw-J+2Wh5d$A_l$?l(*Jb* z7tqbNe75*!S_^Sx&_+X|MoWgYR#J&vDZVJm&{}IBYMaTy zDUB?>WW$)>^i!?M@X`qy@~n~a>=YxEkcdhAe*Q}3E=ZiNw6k*`9uRD*+(1@Kj9!wA zUxF!2j9P1bK=oAE!Kt^DhYAew{O{F4c`)F#c*GjzJYMp$ue$7stwK zvTvH->w$Hp4MtST3s*ZOL8mX~KlqwT``zGxbvF&&a-OS_#3cx&jEm6;P6FZ$MP8DK zxrURm;>wtEf@Mxa&=_e1nsE(&j7`|Xb2zrC&33m)$ab`4>J+yTJLI=vK69jY(!P}j zqB)C^#fzv|?&Z{>u%BAB>_@)sXhJn<`_>kls zG)ydKL+u%s=wZYI=CfY}iL`?akP9vx=F8WA1SyU!W}8^la=wC4*M$|b5w2T0dGe;H zoNQ@oiavN)**~z|>~qVBY&&;|K}&#A6@Si<#G5T0saz2!e(fty5hQmW;`!sATg=6X zq%Sy&Tfuzfx`%C=Ggj{8~ZV4Bs&9EBbTfyh!WNm*kdJGd%kYY^?er zVysl}eTWywIetSbSN)dXcrw;AN~IwPB@R7lpHI{>V+L(2&CR@c<1fDwY~LM5eG-3? zVMZ~N$!O9r#s?chWb6>SUFmBkioa-;IW69oAz7pbZ^Rk!CKnR+P_tz{mD5=EydX*Z zV*>U?o~nqQ7ePj4@B)smGZC8$c54kK2;X{3jp`KWO*(EA7FM7jGDTz`IILmT_> zSEZ)jlYMKOQ!3#@Y_2^H>0PDDlvo!(;>=C>6YGK6rR+d^_A5fHV2{k#;$FT3$F zxuv&kX)MWm`Dt#|Clg9h*4~9tRrg8=M6A{Df2{O?uO_f5S#a=?*Yzb1(7zJVIg{(Z z&jPxRRQe-s!k-jY7lh!k8B<0xDIlfzuq-R3cPuK7&nRvx*7-O*O{HE}L*BlJ>0icO zq4T1T2G?rl)JJC7E`~=$V4fCz&DnQ)f>uk z&hmbLRL;YUceF0={cqQBV_xpF7N;wL3O3{T0D{>z3*9b!l_-edWZ1y#2PVrXnPls3 zr=)F-In$!>HR7)@`Xm#5yUeD7z3uIPq4%nvb_47 zj@^1~dyM)kf8WTbsP&vn^^d84BH#kjeDjd!(N$h%Js*x(@etD@uZtaV!vZ>^w`dUX zb1`~{S!$~73$Qz^EZ%!iYf4^Rp+pk445CbjMf72Yk75@{ z>Zia0KzYBL_uvrpS#WzoBVL$4RE&Slf#Gn|6LF(p+^T9?W@a%mR-DK!e;2>*f}a*I zb`E(TFK;8#!(a(Q@<~wCTR>oRE~-s258?3p(QB{K_2SgmMTnn5oF1%;8)zDkSe14A z$2<?}jvRrsTHp+h~^LkSeCgn7=BDrf( zZh!!o7%mbbKwRy~3cEn>uX~Z9>^q$9IMGf>M5IC}4i<_+eOm{@@VUbi(OTO!Q}Xs| zVB5QZ{!dh5Wl%V)xE_iZ{ogEb@#O-Xq5eCTHU`m1I77VYbV zVk2Hb9^?V}kdRG;H2X9}kIns4hxxL`8Q*8(tGQ#F)L!mRq@?4ePFCuj^el0|o~*Elj6rT@;tG-@6We{zCWm^6T7KjDdSh$QF71?+M0c z7mTZPtPXZpg}#JjZ+;$MF~`nv{LR9dxcfFULxMMj)wz;9)8MgjdT?U=@!#@eyx@`# zG7wL7!FK~C zeFXgld5O4Gwz2cPEf*c#5iD+kGAZzkJHj(e|(ksrLNp z4>}LR-l`qR9CJBA4?pOSb{$ z7Kc9!!<-q3;?vjp%)jX9i|E6`d$6Ca=f#a>KHISj;;w(`{PXXG+RlDFjDaxDY5Bsl zWh>r{-(Lxsvspe3*=fNK#yyb5A>6|4$<;uX9E~dZcc%#0<$sO^Uh%e^kBaKoT1QKr zzk5Q4k|ZiB+LLg)_oZrZIq)QmDvHrpWQpy!AW{|B1s16d>9mgCFoNkDn-Z~|p z`v^&?qfR54AqsqONmj*oF?ow+QcxLE`xAe7z?pCBYDccEY0q}SxQ${}LO$^@Bowp% z=e0LvirlXYDiINq9^G5ZnhYi@Qmg8l=!&~%yI%}sN?{=;^E)zD6eZp(N3#q1vo z5OL^grtL20zP&OC*>;+*U#)uBs{%*mwQ!C$VMY?tuW!E=$tU^V z*XgBhr&_(4eZ_osD!qud6MYe9KST^d`H{|1`%cBnkg8HFnx8hypHr8tTp#_Mj{LPaFyO z9Y}q{hVN5SXdPD2$H=P{aLyXqF4;%I4qC=`vp;u)6J)3dJ1$o^;+##b$dTI%!-V4$ z&R}Hr^f0~cmsSn~TH|La?=*@DzGIJOnP1HCsY}9)DB^s;ha3dqdUW`|}@>o+8QM4H4)YUQZwGjy$87gDdwPNJd8ekbpsfNY(X zqHd=%?HC)Om|^Fua;r;_k5U6tABFuIP`k<*6Y=%}{z$4yGOR@@|tn?+_cwl3LIT4`S+>La)dWrL%S!(OApbj9boH?EJof87$jM$Nf9#P9F z{EKiwhHN~yVcC)?{Sf-{Y)6s<&@tAi4i68t*fzSO@f$sNlb>g?varwb3d z*7?TYua9Q2`+K}X5%xFSPs?WJGC}aQrx&jplToQaq%vI&`y#IQEDq(49luw=0wHsj zJ|;j;%d~mMD9*dGH6@L}r4*8kw##kQJMzJ(=6issCACVe{tm45{oQAf9p?0jOveLg z%52jSWoq56oDW-;f**@_NeR|lW@|xi=YCc9hh13PycvNfQ9wB6 zAkBSJs_F}t!bhA{$36qkcOp(Z6v^D7E2{fHKc4aHYc)O}R5d$li)-sYS+cPkobwn^ zqdv`>-u%!RoW)vi?y%&Ja#t8u%dNWo+KlBR{n(daG1E=YCPz4!%h(326j8E{~EoNX&IV9CcPHjpc*RQJ?NHnK7Ei{}n(7;e?x<_1)-nk7#4yw==c zkrpDr5B@{mPs<8EORj0zkhKc@eOG1$N9Qa7CN+$Ao12}W&S>~T3DLLoI(WeQe)3t8 z+J-Jo4|L_McN~AeqbE1E@s}HA{M&eRIhkZC;oGmeA!SkszBaoopdu7JWRm_Y4z*+) zz5kx`!$4|A>G#lQdxI?oGn&4v&&*AOY<5GnNWPluwvl?WT(W2&M`d0DpdkO~Xi zq0$U^?K;TRZbld3wC==s74px7_Da!``jHjG(4B~!8{Z#~^sN>4B)|JA@ggxS9!+!F zBoduQOBI;k%w=86+X={q1kL5fSvCr*CH>e=zRJnSi-FkGWn)UZK^%@VvSQ{t0dXG< z&KVp;b~ePRLbA)47l4-~=mMoWQ_Ta+LqD91fIdkBgxLlAIGKu6+gz%Yb`PX}D?N_;4!-D#^e z2iT(OS0VHCKD>U$Yn+`i$%Ojawa^}263b`P`8zd8XZ_m|;1&8|y8}2u61nY7O{P9# z&gEyHI;&RwqyGD-pfOCpd>Vp`1_xs-cuuMTfx5J`G-80na+Dyk3}0-ds-8wJQmv^W zV-}JiW>Vze2;Y=b#fS5dsLxIL@>auF=0Ba?38tAp%_XtfW2LceF>b}s?8U$m^d}Za zA*`eNwP@JH-p}V>v1 zdfj36d=mYYs~bWLM(qQC9)L+*#bI&OhXAlBJ*I&A{+YEl?#9jOIsiLnh&uXs7BhD}tFG^W@GaGAa(6~ENHK<#G zohw!^FVYwP-~E0C?)Sfc|DFH`fjk?O4LUI~L8EE4B|17b_DFuuu-tOgyw7pU=CQQv zgRJ?#9{`pWoE=a(|ICVL?MTS765X$;)&2>85&LhJdbClMWtRj|BAW+yotvT62prwv zET&3s^E|~(LyHw)XCDr9e>A4VqDJ{e?kw}xg<{Wtk50iIh9^ZZE*;<9&w~|`dZYZ} zgCH4y6aS~-a+$W5K(J2lszIz_6ZCBoi{V%oa5G%ycXDUfV?gv%!N>$dHYaS=GxdcW z1LHza1k+%8fCpdBtDR+d+5p&&X1ilq2dY=A+dP6Yr3Pz#_Z+8%&$R-hX~wfNGK9?d z2B({ps!Spc#M3Qw6Lh}9fDpt%(?kFt(;6#D()(e2bU(UKe1c{7w;|GqJ*Geb`wW9F zs%V!Z^>j)skPzx@>K1nE#jrd&ZZw`fOEkodby-8T8QrhpU)8Jpl5DgJ3h zKr~7r{qYhf-{TG?H66;{)72%%R6Dp5)BG%5#{tx*9oIZC_>@?zL~c>J;?3f?h| z`*7<(tsL~c-aN7QWEevR-%>9#dF@<57MLyWuRCbO+YD%`^fKZ>4*_s?4k~V1S{f$I z;|290^~DPu!Th9OYK^9V5!rSU!Kq04L#Gy0S<2>UJs(tAlh~z>8rXeIqWnS|p^Y(C z>4&4vVv;wLIleCFL0I?X>BN$F+-RobSL-MdZ2W`s zP5cbK_@akKywh&dd-~X@M>{3fJ&%n_?|QBE+mX(!?IDYi??DsBYe^7ZKlt!oSZ(9x zY;y=snK~U;pi<(?TH3Cy<%uT1U-0+y+gJu%j7}iRWF-+0s`k#p)jz z*p);%Vsu$UapoAxD^LBUD@FN5aOZkf7nTz2_=kP_3EJi@d!0g-TgxG_dVLN-Y-m%b zP>2!<*-Ge4+2^y*)%BhfX+)J$5;3^gvztXZ6?QG8jAqlxRYu7B-@aL5dyA7D3)ua2 z2hR3ZG}WVZJ+KxM;Oy&Kt=1>vgz>;R>!b+~T|a53IXLt{3`tk2>X9f-hb z`8jnoogKo1(Z6+{*k|ef(DctO`&c{{6jP$=Ul;{ZKPh{Wg}rd`mxIGFbcE1|Yq}kw zvrTa#xH2T9VRk~ZA%z5d;k-p(D*46Xf%LQ{qOj(xzjl&$B&GqUay7izt+ymFhT)H7 zpBBUxoy9=9ee-bL&iQOnhqm9CllJGr72IPG_7A9+9))3dib>b`j16d~Mo@Af4@U%@% z6fItSK0~`ag0DfR@(%=s?kk$I+O~sF6-FdgAoy5+sh&Rc4qnW8sVk+vAq{82mL%sQ zu56G&ALad#Ez(LpVT_|#+NQu@LHkC_qRH;9olcME-aN|3S|LQTY8<{1v9WMG7sj_6T(`A!eE`gTjsy zCLsFRc6-g%+8`B|Fa6f9IbKhc>;(*!h=hcv1~{u3cW#Gs{2}g@DF%$hCZ(LWqoT(I zEA0PGt!hAt!Qh$iUt&g17k_ZMmJ1+j^u{Kq433hZc}omR(#kVl+%>$nc2FuIz+6A z0_hsWF45b${_3cEZU_{J>5`e>GR%h$2rm=_`S{T3mVx=L5&Ubq&%cO0xQfSUZZXdLZ`!Y{%j(r@@;`i6-Gi*SXy)=wo!{(xP%zIAd zK>-M$xHxdE0whg%3{GqUOVKYDetP`&*#AsiI(>W#&Ihi`L~ugL)Wx0S$PhSM=?q&_ z2?U4~dbw<7$F+@Lv(kS4vcS*mey)B=oZ~19)WceuFtXqSFZ`sRLUUe<2u+N?!x}&I zQ)CD0eH%lKJIr@8j&Z?_yJbFs!unE>^%f#r+1LoLiKDNY{`GroE76O#&qs=}nhlEj zuLnLa)V!TZ$+3|sD=z|^g5M5bG8@alvl-u~W!S=4)xXmucb14tFz>5=W-Z!fj?-n8 zhPfqL8c`_HP70z@kdTg3zEaR=M0?A2P98f0kzb3``cd-HB!rXdo4aj+NCPTioAS+j ztmh`tGG2gMxF{_uI4HP^y9VBNA$o@1f@p_SS|(nzU$`_Ll`? z24z>f3J*0hsjZc!<%g>Vo==I@w+(=+r`07qmBq^r1;oE*oh~a+pYyGe8&JPuZ;Ffb zE65x@gnLoCm8FAiMH=W;(iFX7$4dcgL5aFgk)QejWVrj0uNm2$j8qEIBZV`VG(-Ok zgo-4xBvC^ge?|RVuj!5*Ya@w0O~gOWL&il8f7WG}Q-+eX|6^ZrTly$+-N*OfD>`uZm+y$Pfz_vei%7kzT!;SAcgB_hdWt_ot6vhUgFyBUm>`xih5(N^MEVJ(IRN5lQ$?*>jq< zo(GJnl_Y6HGSO8iD#~g33{|L$@12cz8iQYpPt);d8{ea3ykn&GSX}MV`=i~00$Lme zBJHSw8~v;jZ-vypr!O?#J_*+UV=k4@2WX|nUv|>sk+uxGR*>#KEzRd3)EZHGVGL$| zWMGGRxMCM|)v3p!_LCPYxEz;jm(kIvy#85~C=mm=Q8EWl%DZUO9|1^;9)|}pV|>RY z@H`8t-(6d6;e{`)gzE}I?!)7(|JaI9CF{St-Q3)8Y_1StJ1rVH7HwhuS!G>U4!h|G z@56u9yFaqp{tlFlrRM zc(IP`v_1sxh953@{y*-f%kn%%0a$dsH-626VMi;`5uvHUpTWQ~Z8G zH;Io`LqA~Cj`e20phn#fqa6_3bCZ>yL^ze}TgFbf2_?!7QcJoS$7B4NHUw@!r z=Q&i^(Lw88x^cf&Cdls&&3)--klFUNOsv26r)nbSWk=*w3Opw+UfCCWQq?wrYi&q8 z*DuY?IDM0tr}I%Wg9D~s>g6O0B>b~VqLiK^Kq(jw6;3GE-U?K6-dJ#BxZw>>bWdu( z-{j1n&Ile(bs2ttV0X%KL|VMvr#ols!)A^UPQ;Gfe-Y}5_UqxI*Qcxr#}6y?8J2M^@BtwPl_ODKg4=T=hj6g>Q&OeSq-BnUm@5qg(8#S3d9@n z`2_?Xre+>I!dlsg1ozl>@R&8Z=Y?C=`)+00(M|Ti-^u8@Fsx~dr?#kzn|oF>8h$;~UyT@n zS;p@wozi7XZj1ZIKn#Z`?1C>gg38|^tDvFd1tJe(>Cl&~v~y-+M3M1?#gQL14>Nuk ze`wLk6u)BXS!iz3j8nLX)L=v!;dsAX7~*!8PI$FQSd<7&zc?YTd|8<%JB0%2H3Aw|7a z0noja%|~B77rkmY=*uk?M;f@GBZ9 zPQLmXa`pDp@u&0)r(rF(y2DRC(TQ)!k-w~sRAq2VUJD!(2A6tgO8X-l6mzL(a|P>l*HD1u&Y)kqkLr+vCnl@? z5#s|^ci0;b9wKdJb~rOBSak(7NC+*!vsvl3sfkM*RUT&Py3 zr>pfI8|AN(5H7~n0JPL-L_j_G<8d6su?SoxPcJsk-~2kOMJAmgPRB*tn_gKdiR{TlrFq^)40mp z>~DGS!K3v=+E1&H0wn9wY?!BrG?touhyVBEwb(7r;zl$qztnfia$21-uv9%Re@KXr z#yVxdi)(2;^Cehn8*iT?Ga5ZPk$MX*FYWoS@m&)l_hs`ZiBoSl1KB~iJ*_q8OveU( zSl+|zCy9om3YGedi68vjg9MVn^)0x3^D|18~o*2C|#rhGQi zZfBKiG8*b>pVpBvV;uR)2GQxs-$X(JOCC=q_K!d9`N>*@6d+r7Jk?he$`P75g-Dd& z5x>dlGY>sr%3*R-eu2l**}e08m@S9d)Os(uTyo2j(iF{aVT*XZyiw}b7k~6;_8LU7 zCuu}sY1+d#|Lw%(KLhY8&_3MKl+7}6v5JOCnIKRw2W5|3JpNz8pb>xzKQbi@h zcJ+y$e^t{KsZy3a=dpgkW;vzxBV$TEIVBc0A#6T?7vFO{)w!cLCC^H4odctp_oJnU z)v9I!R$|I;=X)i!vtE2M`fBI=zw1X(I+zB>$2gDPuU+@aa_Dqnilq}_sB1rYEhD}Y zK{mQbkH@|0F#8`AR1ryVeUTAm5=<&1nxwnv0B)ENkITwsr(+N>l9u< zL3FnoFDm-KtBYXmJc+zPkU;Zeo?Lcza0jUzL}O}@SE}!piHB!*FbQWZrmph*?X2$t_|R@ zUsJP8>m5npgM5^F(f^+hvO$h>2KIbGX)CIdpl^(0S2r5UK`_EucCD~|jNt0eIUIWa zdGpsvB*lV7)~KU-J`!89j^O}xGgn`Ys>SA{c%W~pBQ`EtU>bS}2Dwf$SL=M2@N5oN}X0cWj74^0@HeLsn_3mFA{Qwn-D-`hn6;96dqO zc|a{IZ2K00(C3;U>7NDy#9HbMqy4m(%m*ZxJ)!~ACf=)|SR6@66bzyeaaG5@&t7cD zaBz4G$~s6@Q}?`v1uO8d{?GOSpDqPW7`N_MlwXU{`O_=@W8~JJtp$PFch)S?X8<{p zJ|_RhjX&W#)zA8r`R*(>W)hFrYS*VyBd4byd0Rb2()8H-4hoylOJkB0>G;vysSbq~ zz=az*zt-H|_TSt8ogfL3x!P@|0};+vT9d2%{J9GpUz}6>8S68t0k+?p0U&%0RBTcQ z-+)4V2?CtOd}mxR`iL9vsB#HQtoZll*sLBNm^7#~(!&^Y)DIlz64;*@p#)|&(M6gF1FMlsuKW($>~d8^tn_d!3QHySbjWUJ>(6lO!Yf zxfo>Dv-CyMX#9O0;Clup#9j@4gI^Gt`M?0Pblvikjo@q}MPm)b`glyaB5yu-@U-@) z>T>x z!^vx0?b?5TcAKFofUIDcXl^&DUf124JlPw~CN&?ckZH~)$oZh^@e=Esvfb)|8!MMM zkM_imo;8f7Ug9Jw%Gdj*e_sLXYn!3!JWd&??)@_|LKHrKE+@;!&u;4iy zTG^&}gdiXh6OlxISl6@kIyHX2U|F18sq<^SkaFWb_YkxHR$L-tNff+9ya1DB#kwwS zIgSfODoC{iV+;8f0(Rn=8T}|~lH#%H=4epS%S$1-i)kf+W4sDrcMNtesI(2jJ zN#_IG@;x2a{Sg!t$u%*i{DhVw;zK`K`3tQb$hV z3RnJ+jrZT%kNyoqUr1d^%7PLv9##{-trjcE`J2avCgD@m53Nm`Edms=f?$r`|6Ha( zyPcM%PM1<{x>~~+d=xmEi#brHOo~206Nc+s(}VR6Wh9zUc-^O-yyHpnc&>A2jXp&} z9Yt3`{ZE>)sIv754*!4sj%9TEp-SUQ$u*YB<2pJhpH~{CL#$@oJbxm;F*#cXC-$wR ze8hq*{KTvZ#*XT{0RoMgAKsjXhKAMF19;`d3W;vlJl{$W0W=Xe`uTHr>0OBUAUNn5 zbL(}mE~b&1g2|tvjnB3@IW=>rw3uM@b&2HwDfk<~@-Wx~W03hxE# z+>J>@o*~W2`4+E3LqJ$0hvTqbfVnvgt>P>-C-5w-vq6f39N!W-A+|23dtgH;JBp%j zU`FY;*Y85%KXIcvWj=P9EMI4s+z-3-KTw=TyPzrjioi0bT+_uauy5BPOSeC z5c^pFD~?24hXx-AsV;i%$wWi2VyfBrsWUe}fdO-mx?t;MgZK&24BD>@rBmQSK8Dff z?~A;u`C%jYw4R(`g>XP7{ha@N=h!zw)=SZyV2mRx{s)Z>Ytxy4J99jUc7qQ zsrF)RWs{l`Q%P3OJ=;;ytO2yiuK20tm1rRwnmjKcb82ACk*k8U5jXxLZUB|5E9mU3 z|5*OVVLP00O8>hZ->H4qQr#GAbN z(rVPVOC4VK!G`06??sbx)FX3ThcBOL80QYe6FAvr&%U1(B%n-Fk+mdGH z)!k}*X}aX{xZ%(@rEfH`F5oBDzlNMj;S2at>S6`P-6Snxakd1*YkL>*C>t-o=tfu0 z=nUeIn2W^!`J>C#GmnfThM>$2S1dl6IG0=o+p(hovLQ1XAhqKbJ(gUDOhXU%^&2-c7SGa@0W)2fpG23#*^<2aXHWhF0%o zGY}eieUe7%RYMY@$D{RiniN-V^!vD(hczia#+b9vl&`&$vE+vKDJJg1zotL4@5lbT zctk*=AVYnDjh~R21Ha2NMBcDK+4NzvZ0P+p`q_ZaC*uU5ZBr028lsc zG+1GGl9_dzEBe7MuZHnF$?1@}1IBBxj;Gr=KU$T_o_`@4jc{^KF<`Sz9qssx@uj-T zm+EuuXKs2i&BG3$Vf zi)gPuF=1U7L;>ocE?|@yODaBqvi9yB>MfT7kM<8uiD&WhkQji%=PW51iJuFu0r5(7 z8T%Ac*|%WC4f9{pk&B@w?DN&9-SSG~y#X%&#K(M`@snsnqEPm}5;w^3^@n&8mp8KB zCiNAnD%;JS7=np$K8IiZ_p{)!=e!DH6PnRDQ@CZI`(;w9Rz?N%H*L^q4-peHb0?V9 z$)`B>-@y{XYbb^A6B0^|_End*UckAl_iwzt%6Aog(;!6Q{mgm_)fm;Tj{~o{gn0;c zC!}fK4-#OF^SOwKquGL3H)K|0n$qyUIuuZ7YZyIC-PL3up8owi>fRD)1OV!zJhha8 zDN*h?MVT!?9oI>~9*Wzch+GgA2F>nHOUbG(2t*c{%&AR8MC36>sX$3Of39 zXgu2OCLSaFH>Zg~Y{Dsb{_9tg#y^*}5Pff)On3Z;FJTP@&*rll%w_+{*D)BOJ?*d5 zi}7iy9pwxl{QdV_$ERQz z5hRMF(HqEAZjJ2h$~F{iJC+Fbj6;Sv+l{q4g=0mPvp6)93@6Gt1$_QBRxxh*A2)|<~f1$9T8r5$ie!b6j%EsSQYuC zcK}Dc>jIm%XTCr-5+#PY1V+aTA$Q3KgLXT?bXt0av#tm%ZjUY1fiTCqf&ZIOkgBxu zH9rPB(H<7NiW!8lGYVSH4;&^VX9h1!H}G$XF`DbWV|6o^k{rxt75P`Y7K==N(!4PMSJA7zM`!et_i#SUgd?riP%xzlh>jel12{G>B z=nyE@mSh_U8M?!=$?6nRt}aUtgEHhROMpypw(Z5W0DffCjIY_%XBdp;FnK=~D|Bt)belSxMVe*c z^V{KgZUHJ+gZ$K4H zT|_7S?xh-<0??1JdSTE76RXB3NG-D|(B65j376XgNIEd^w*~p8EpWYDn&F@ljWd z)Bn|DWTlaK-t@Y(UP8ezTeB4q-e+)Iy_4d8(*vd-rH`dDYuBD(4zj~>(*b{X%i#VA zaMq4-I+a6r8X&cX8^KS69Qew}s=vEodGQ~~Xt(U99(q+(repQp=eUY}^&w4r3Can{ zO`v>ND1H>0`2cyiyj$_?;ReU;2DfOP)mjHQd2p=2=P2+2ewdrY#~Yke(n+HL5KXKt zc#QQ_7OCVa8&ejnNUm;S4E8IYrna)3y&?K#`v~S5G(wbfwn>1#XG$^LMfDs*cU^Qzut)} zCI}YN&W1b}p(f$LKB*;EoOxQrrymn7m%>#QAi?mCLD^u}n&wF{#-ww-@%_rRafALR zaIbUxeosZw2L_i0WnyM8*nq^bnz_@kPH78^Bd_?}{m)T};@r#(!E*`TVUy zYcNl1uQY2^t*h{V5rGJaW*M<2D<&PXBK;z#&DWiw+m&shIyDQ>h-s?slUzP^)BiuTT;^~~nrE!a5Ljdq#P-gViOiF7gKbXPy5K3egpgCvkIWBi5UXbw!ym7n| zH!H{?yYtRJKpi)Tcuw%CUI~Hk@$B15pugWyu{H)Q>=pA=QUBHQG!Sq3>S>rjubPBY9~NFcTswI& zGm)YqWs^_2`P6&yO}Z3o*xKLPUpE(L5}*nJPxEd3Q(cYUUQW3eF|xT*A6#veGt{+T zOsFKk57WYeayP9fh^mK}$%vN_0i^9Cqp53P5Rhf4kcvuzG8YxA0~vDfk_K`g0d`qZ z9lpI1-@3jXMS?e_tWm8RHndJw>&jQl6YsOkI&OVobveFzrc(e(@{f8P+@znYFhe8J z_o{x|%H9sWyJ8a}majqzYJuSBg@JB^C^0-qaFK&pQ`Hw%Hk6jiiw($sQEhcMG0Ev? z4}3V!!&<{Jf(+Ab>jp-||@E5qO3z&TO~$x*5@=JUWOLW|WTDevf3sZ?=s`e7%*@mEOY%q9;I_~akJ7XQucM}*HDt;H9QHrKdAY2M zlohUv#E6(F$^0#!E_9aj@1At&boyD$^5erb9$WtbL8^X+BlLd239rpy*cv_$v*00v z(h>EN`yaV4)Wq!{w7Xz>N3)c*xXnGHC|ckWfuRa(WdU z9IUE@67mPL+1?|o`iBjXSMEaedQKlbA=zc$uz@3VFR3X|shYrJXxmb06EDt#p-U+O zC3(`p#h=p0Uy?8h%d;>(`6mc5=#gLMFY17}^8}ROoIbR2lZ~;RGfxJi>fh~|l(d9y zR+dxS=2MDyUl8NR?J@&liJhuXX@f7AAiHK!`X|HGX3HHe0Hv=^S^*~XdnvII286L} zbXDzZ_&^H0;l4M!(xp!`C*qf&&PA=_m6RJx*e^Sl$zb`986*rD@7M_sy6bzt&MZ`u zFp6j>Cag{7P@yIp{IZhnL8>5bz$5(Q)2M`1ZvD~S8nS_uYgTmli!e!HK8jE`2!s`h z-W)I@qR3St#7EZanv%_rtO`_m!_!+RHk8Sy-B*N3yYZ!0H(>M#^eu16vkOHt(7ZXL z_Tg2V2UjvDm#PMrcsq&(Dn_MIgL++0>f0B3_i&s~M`lva4|2&s^7g+|RX{}ngw`d- z0^i(lyZnvBaNLby2K3NYgD+g|W8z<#zq*D(Nf+^b&UeU!CcN~;`YuTGj^ZPX!&`od zq>7X@I__;|5N5KPZakM4DWKfw0@IOBKY?dh|3s*s2Y-g|pjKn_#81U9CeQy!SnPb5 zcc#`Rw%E)n4MSM~=octWLCY(yPpSo8lX$oSB|Y?MV~;059>NP(>>5aIEo1aAN{3VK zfV}^hK&q53tTt&0irgQ*VLkChtTfgTHjQK;RgcJMPHX#pJ@!f*1fmM|aXA-S6$WK! z*3K%p!iO;&w+C)^R!X-#I*=W?dU`b@jbQS45xnB*KTz4wsncf2?k-a~M7|4YE5IBz zy^!XZo#M?|W-*K36N+f*@wc|@E5esBW}aQuqI1^F-#3N1rBR3*g)|C043MzVq>*Co zZE4C+B|FmKy?U{&}@NclBdA=m6&M*55+`+iSw$LYoY#H&Pq&tz6eu{xS z`bm@<82AjJm1_GIbY?-w8P#nGu>Uu3{yaqCe#3Q3z~_t$eQ+{^1%~6f&^PzDtcAGN z__r5jPub;uO@l9|h!#b$nHhbKOSK5{cZ1swe{V?`AQ8qiTV;R6IHKNF(TP3d(Mc zRR4AR!Ikj{7*Lx7kbr5o=%z zR*RpM{Ph8ef9v*c2N^4oME9*QT5m4lMby~{(w4;NYew+_%zisc>bY1Pd=zWR=qGKDLNI8M8NFY1svan&5v>O~na0fpuwJDD7$#Cwu&pF>uEr*B;Pv6DNR+aZPpE+a7RT zsJ-$K4~wL}e+=@0zDIE`wRv}}~M6Y0_+YH>6*C+m*@JyBzmNc(I!CNcaeg6+`|4QDgy%Y#Y#r)xs;a<=F(63+kBRP{Pe-{o z{<{KjMu1(cO~t6<^P~P_3jyN|nH2xjw+|^PN|wZ_O@Ze>-xlrAPjzT9c4Z^*YM&r= zfR5LsPocr1{s_<)-9Sq5={%kI5iE~=(}GM2<3y`uj}=p(C5yA>dx`3OavR);dcS&5 zHxZ}TRWuOkaHciv-PpTw%%+!6J96CNF5e%UmRj4yDfuYdOz<(ZFB}kx$#Uvu>(QL5 z4Qi2e=0U(oJMVc8qy7hv)8yhH9*>4CWo?i2CDdXT3>Hm?P|V%g4}ZhQ*ww=5f{1qu zH1^~_2G;BHh5?J#XN-^Oe?5SR(f+{o$R%9sv?y$tW@O{+^1Ce))cU zktK+Q+TXLqm+0%ucHx2JVz0=@IAMv#<R8Clod`04d$vQ6(KIIv$ z7*leBmdtca3dVcfEYR-kFr(K-o68woaNO?a%W&ngqyYP26nFKA=laE;Qho-bEnc*l zW$yc)=rNOv%b$!tjOGK+z#X}wkl8Z0+DJ#|N#-)}1C0PT^l9g9l~;+} z#r|)col^tcvci9gd2C=&*3k;J^KDWA{i2;)AffY82-_)=Z%|CszGmje4xqKbji zpxDQvVq90Ff4lHelaZa@+}#zU_a`4kQ7cpbbPti1th(F?jdm2aB?&pcs{CY1%s7rc zX-r!RgI^1IHc#?B$cH*Uf%9>Z0~ZngQiXo2CPyZyeGBG+x_0$#(9r`HLsdU8V6IxgDOXhtan zDK_xUoVXLUSO`fix)IfhS66uvoIO31HC3duDogi?zRbhqBzGJpH&pN~r3GBPwbmtL z%KV3FZRA2`FD~5J>`XsnDMSQ+-VJ@xCf(VwMb;bZhCN>-ECbNadBOSX&0`S6Hl*RA zb|-{1i!*e11L7a_1>o3?dgO5Ik9g_U z=#s@Y*8@Q}irBDSPX9Bv*(Z}GQivy3XIf1cUR&jI{S*>4Bch4QqiNt>O{dcVx~9JWg-g_dj}68juFh zbh^^7*KLy1NE1a!dH#5SI|15wob~o4Hn5QdQ0)mi5tqGULPH+~m6#Rqha+@PRJwOdJtO zD(o+KM}=`d-O1Qq<4z-f!7wCRwK2GGpd);@JP{Y?5Zu` z^gXj(9oTYr zk9v?mV$eN_x-oe?IxI9JH-6cq+M`3ldTc#*Q60Qnf@yg5-lJr^lpx27NRAWb$jE>$ z(!y&RX;9xl^&__=7o+94W?NAnLg>|V`9$p7laKR{P**t$3#L}d)#NAXBgbqA2W`Z0 zr<+dnnaLxBWE6)dH~?2v7K*kGC=&MaqlYKT#=+4Iq}6p{-#ro8xqK?9;GYCly1_y29<3%3vKR7qS=s78Z|FQGZ(;)2Kf0X8g`W9vJ*dU* z*?^9-4(%bd<}XNQv;g6a7Y}89Un-kbRQS#@23s|CdQ_pp4$D|`ng6fy4=e>_hQ@T) zJ8cV%4nl0NSnVNGzTpIIC9+seP+iTC#auhs$d42~5xKK7B>KiI2h{-zT*7XCyV zmj_&G-c*vQxI7GGp5PgL8Pk#(dFsS<=)eDU_BW8z5hC!>P+ryNvhR=7j)m_5^YHJh z-71C@71&**(}j_Hs6BJrYF*<&*<0`&&z%cuCi8LOQ&<>u;W+S zQ~c)1`dB)@JY+TjGh=ruX2c(Xcl@QYlDzuB$@IP_NrsyYe)5W|p7lWlfNRa-!|y&# zFPP8}=lWfNDF7v;fet#%V^Ef!RwU<6k;(T}5M+m>A)U#x(>U9@0Ygy)Dr5hRS=( z!92isTndsekN_e)6V=o<$27eA5u-k*99@g0kPC%F#6=8dQjyii{-=gKlu$~nt~hmt zW=D3bPup8tr_n!CQZ&P|pBo#S2q93vHde~8{RbWOgD<1Z2LA+tMYK0Yh5oTdMBXC; zEgJOo#{%Wf4J+iD1nwPbBA@k{3?`LnTpgcXN41Ilz|@1j%O~hr6J!%$`z*uc6sEZxplXN_c^5$K}(y5DV2q zfg;LNVV%G(_sNI`K(sEr-@HHKfNb7V;@3wub@_gBB_w*vI9EgCeZm8ABTls?a88X5 z@fOVqeS^;Y&-k+=`|qxd_pGQ`inMO-oDhQP&gMG{L#Pzuj=L<_9|KxhQPBqs^Ihc> zxa3@&wYI}I#@?~@x={KjsABIP$s_yX;cqxR>Wk&?h<{EOP7+CIzkv4TmhczMkn_2! z-;z!qz*Xqy*+Y!Mi8uLSOErQBXCkcJVY&YL_;!UZ6M)A zUn{_GM>}i>lwJTTMF4KS#Rn2@zgy(4UJcK7h9I5Lmk{kN#X;9t3~~mL7LR7AM)CFCu;#OruXuHZz~Pa#3;i;nj+T?XJ69_v7*!!RF2B0eBJu<-SN3^Cv(=Ks$p?JYr?{MHQR)bS&m z)#K?_di@i9gQ%Gsyzhbo7Lo|Gi@!|Yl6lahi=U9c#6OWRkh0TDiS&%fhB;K7-OfdWmO!Y;WK zG^l8?U|noMYU+@n-1)~=1Et1_kCa%IBY=|#FD!_Eoj%oEG;2H@2Fu+(!W2ge_w(tS zU0C2~O32F0{5~=9IC$h`V4~8q&)aJ0e8@#bMOc4vr2By@U_zo3gJ~4~Xjt)gw-J4WZWpk^K!`NP|?h_C6K9J3T`lQY~)y=U_K_$Xgb} zv6skV35pdTn~T9~_l*oLsDa%IsjK%?Q^WTLeXBs*9NqeReK1*xiTJk^3$2b_4Uj@U zBoW$BOxRC%d{wjr@8g7seMxZ5v`)2!&iBC;W)Y-<{l^4>4m&J@PVPY)Ct5)BeX@mD z?qYrh8Tj=y=rjF#NQU05KCaEfQ4u7-2I)aXx2u=~EnUlj*G}c;9rO=|IRbY{`f(Us z{uANQ&=6YexY#DG2>?K}LoXqg+<~F`F!GQ85q0_c!%NA95=< z{mGcJUaL~p1m~4BurmZ?un)_$Ry{9}B_)gaB`RM+UA8w(utD zYv?hK+~4Uk9W6t{C<{I|BO8>H#3zcb7k@30ZS`kcK!OY(AiEQd=PuJ>_Q7X5Q2LL^ zkq%zG!Dd6GKCX@@PEUi}fMQiv=QB3lp!bC0LNyUS=_f9O5XKft?*%!IKiz7J8BXhs zdveIW=`266_KJ7OSGyayNGZ_K;QPhg)x_MU#4-M@Gj>z0hR{N4I2ZZUcPeZ8^=~N0 zSR<@I|D^r5#`>k0RD`_zQUf3eQ}k?aK8%c0{}%{G8Zdm76QP{U5*3*)5Cd$#j`Ow5 zyS6OQ6%b7E?P}+XmagtomKb}WCy~i`UR(5*(a%-YTM2XW{URNpTOPa;k&BNC9?t}) zWXXNU$rsR$D(lnHGS!X4DM9J-MAIJ%d7H55EvYm%pa6CvMO)V^&(8#X3d(MjT{HMt zPn|-^Tv8foGG?EI)%A2OMN1YHuNpmC+IkvHzrJPDYvl&?a>Y4b#;AaqsSP>-`c~ri z=OC|b^#K>JIwALr`Nv`m&KnEB8n$HFgTAw?ffkKb51xYRx(9Kv9u`go&wxG$xFhi-!ec-?P@ z?bm<7g=}r|1#h27$EBKTuB72F(b)o7O$bw+E8^k$nr0qI^`J4m+ipb?Rn~oT>}#OgPDvKx zrqow9eBO$%%dwQH7TnW(@T*wcUs)}`f7XrsOk0uA{^z5XeMq_It3Xr%2Awc6r8KT* zwVwPas~m}&Wf;Ep8&S?*55e^XXf#1FB+}(<$DI(c+DMuB{`tU|Wx>&S;II@+k%9hf zrqVP9P_aZH_GG{u#819>7vq0Yf0>6-;!;K&(^dWZyHiggWL-2ORJIp&e{QiEH;?xI z855a<$iBmxACgVTx+nE1OF}~Yz_Ht~tA3x#H-4Hy-37m#oa9xjt@wNwV}0mBR@-yy zH>ikD7D$Oun(VWj>a50|_`H#EtH)Hn=wx@!o;Z_an-HQ*JXT$%WihZ(iQ*2}{l{E< z0m*>1Q8KsN8a_Zy%*@&Mw*7<-yLFd+D<0Yfkya59cAN_aRz}Jd0M~13Yv;?t?lyB8 zXqj*+;TAt?px@=>cSMDAmV)tFh2QIT4_kGuJ4@|GwduMzgri!HaOOqEuvW-E==76Z zUPyi(Yy1U@pky;8eOswcehv82Ma^No4Ge)!JvX3Jlp+uad76jvZ{0+>2S0o@bwr`| z#-w+N7)@UG8xOK!*y*uPGnww}gLLE}&=A=Nl!ddK!M2$bHT*#@7TGwTph0-gYPk^A zh~nW9P-PrV>u4yi!f9Mui8k|s0)>BM1x+Qyr|!s(9l8kw)8b(e)-6<5-Qqe*H$B*3 zaY@g~9a&uzOvYm**~PIr`zw00e>b^mYcVCWqp#o4-|;WDm~3k@Q9onS!z}UQET0L0 zexC9(PScPrRQ+H-?<)Vnb(s8#Sv_Tq4dEyFlb^Z-$LTqr3l9EMyJUcmEC%1*+4&3V z45A-pP_dT+q2s1q&Bsk2aPD6KcTZt3f;r8Eah|6Xxbt{+!e3)t@JlV(n*+m;A6lK! zs1hNo&cC_Ida6b0hJ2w34!Uzx#73y$?-UE!KDl!0M(RFivEN4kQNeRXgy(|74oaut8||5pn@4=@VA4AFyt zFuRhSQVBjV-pdyUXDgqa4RKbzuk-;TQQ~?3FzOM(#C zB$5n1u0Os`64%L@q8?qToyAaSr`W9gz+uy!(5y)qa3f=a2-|-$#J&sK$YL6p#WWB_ zMp@by(>Rbr(I$NC;+|vdMuDnTMk__W7DwS3UGr$|MwFv(V9gWm<_>ZuI;WEN(CD6J zQ+7X0jEyKxiiP%9z>FgtoUPuCpW4lx5oxo{*jw(lwYM$>B%8_xLI`WW%=qR49RLK` zfH)$nnsg;jwj92gkky0Yk<`P{!8k`rN0Ug{M%}nH&6}BWErQy6pg6d8KweWTnyKLr zKrv>o5p5#lj!JF!g`46QebVMCH+T;`lyX|lcu{DAhNYNIdGZBc@V}#&c_Bt<>$$^e zX;z^$q5FOo%C>EG;QOxK1!LhG$LZwGjMtqP=#`%$j4@VXnfjk`^0$FjR38#+Sf#Z6 ze^Z?b`g$xzhTd^0r*P-Q%Iwhp`hT89E-iN7x^-|_s~vhAyiu5Ct^{hOGTcgO*x66lFCfk$eVeofS(yxrGQj^?rPH^a?*!8`DP#_K45OTnh4XF zKth8l$1xp83v^=!7_D(y(@SV%I?u-ft(fu9XQ?c%bYRpQGY{=yu7Ycg1gChVWl_suT=ns)sHYP4`u1w{cw&#OR9*i>9^>9I z`4B_#!#6_hgW5)ygc|REg%POOO4|)uSbzf9bed2q6}bbk&Frv;p7qIh*RML3AKEoR zFiIaF+!EMzim@a&&*By>^C)=DVzZo^v4>N+NqB2l6p-huik=snrbUh_h>|{AI_S)8 zFugdvfMstbzYcNeTzwFuGJtQ&sgf>d?~J zVbHxFHX+lTz%_caT3Q;F_vL#-vV4g%dX6>)Al}}HI4A;_ z4bv$j~ zn}{FD1ORQbby1?rn!?SAB?6e0Cw>cCYYaQ@IWI*NKO;%nX42lKJtTJwS^KO+wvjU0 z*E;FTMxe_`)!5p!2BWX`Hvz|(o9?@IjS{)qh>xhxLfJ*j)~b!qP@FJBJTbZKTn9V% z+0|p(gKLi{U;UaZFr%rY3#BPXQ`uxReM*wEP25XNK1n4y_A?I#j|^1bqQCmjeh9l za2sZ%1l!80*5je_-rt_G`Zp5Dm-~$?{S5aPipVy_sn%praKKvq{8{&tLgPOfpaYXJ zh+-DoLjTu-Oo82k${l4l?K%BRX_4q@FLiGgZGC8Itge;O#8X`gHMugK5DRg%LKLME zDWK>8co37K?)lH3z#r#GOB`?gm{f@|TzKwC+A-66rL6jq>c^y#Y>MBIyySW z4EF_(SDim1k)ncM>frH_YxGX7D^>QhcW@GudS}E?ZsmnnC@r$SROJb zQwu|)AvYuQl1Uy`Tn8-@Oed6=GfZ263UWi!8L;r0>E0=gAd^9!MeqoCEc~t9BpOrw zYr?M3Z$FaxI3X$(m2be!hIk3_tWfTU(%c7>{z0ZL1GzJStdlXClXqa5xyCvtf>a;q zRB*2+VLJyV!03?>v)>>HPv7QIdAK5IB;O*kV{fsTc~4>HBa3v_u;@%uYWvIhVOfZz zCQ}R~?=IF*`cj-?8Dz}X=}%R4Bt}L^KqgqqP(ZlplNvr*m`PC2JA9O#2ifJgz#r1^ zH}t5DP4*LvmyPVszu!jx&(_~4L6)SJ;n4XA&bYL9lfw68c{W+t1lGZ1`+-pY9n~>K z{QQd#2C+9LG=uc(I|phqSOp`lM4l$1W%_7I0~V~*!9E4+FEFOMn)o3fpl8a}tk4EJ z2w~`O?SwizAx9)lIg~?GtF>QQB=(s=obFJzU`cmPM&9xwEyi|+BqHbezlo~rUZn-& zbEyU{R}P4T{hQqpJ2NU{#j8IB{GYG`4b_mvbd(d|%3I7ffrox8B$Pt3K23HkObpWg z1_B!2lT~?-2|(05udlq({hzt>;ty{SJF?64JxH$n$@pfg>?1}mXdqfb{&Pkfl+v%AZoQpsplD7whJRX_({nae99t|^vV-JAh zyw0LD&9$`|vr68Spfw3w?H z2EpvX^)%2SFE}G*37w`wm*-lNK+PKwF&C5Oa;FUzkwHnsguT*P*KUcWd)qtm$-R*d zYa!s$^^?7(qvKD+AscEvwZc$uGs|{PX-0L2Cw9r#8&uv)oDq6`+8oth_^-?X=`!<8 zW!G%`nb1-H*}6lKe+b_d2+fKa$OygxiZGMY_S+1LaFd-3?Clf}%}7Up5F_ar-vITa zGP@%z0)$w>9&$cWTEmx8Q^`YK9c29)_G)tV1OBRHN7_BkqQ}#0g6|W=5^r@c1vRo- zzmWhcCDO}iXH5M+j|=syS}4W)=4!#q_uMER{cbkKtzyNNw$IK{5Y=T%Tbcq0|BVk->cGY(QmQM@sy-vo7e=0BC3^g?Hz=$ zLHnG6a{wR?GN-z{h{esmoIg6a+w4;bPW7V*gH9hj-llcu$Ah}ACS>|wX{g&-FM|{yzVtcx-{W1pY`wbdbI(1;7_+Z zJF2s?=R=h#DJ2lJ874P=j_*;0je+{BgjKK0BF=ajMc`|D5D@H4-cW z^TZxfMFo|kEI$s|{CsxGN_tE2CR_fb8eh>BzGYdgiV*@?m3W@gpD5{b~RLO%PO+Y>L}xf{KJp*!ki1D&(b?%n*{4 zvQ1kcFZaTp2y>Do36?8EN4{*oK1c=)yT`oujKO<~EVA6k#x8BlvJt1xK@)>&xNX@j z*M{86yfQ&82>6;)UBs(EA5q{Wb??+rRAgK$1JI?>@b1j`-uJ)lWXg4ahO%w`Y%k>p zzLxzOY-Dl4ZGn^lH-7mNU?y`hXJ^8yS;BO{clDw1NXj>HwCx>i-&NEgoftQiL8t}K z+xinBIi963)(|1q@Q9mg6H43gDPl5tq@1`|6Dv98td)p*H^1s4%SA<5&yKc!c=og%x`NK^tnHhQ;OrJdnD)n4#Om{7*18QfBl+V z)U4Kqosoz)$K|}w9*D$3**I`MK|)Vq(PV{g>2^*asP2Jg_C8o*r?-Z=V3$4R5>(_< z z>`0NSi--_nB)kUXz8^tOSi4rnyZ67#ml$&0{NKOD<(CM#cRdTu1RW0{Y#vqre!pn7 zU^AUYc^#5i1Lc7*h`CcPHo7Qyc;MQNWityg!cc9*t+5d3M1`+V<}Q+#U!JUf*Lnx# zJ&LE3s0Iel%ba5zSR%WN#l$1KkdK%%>$C^F5D)XEfhZEzS|oKb%#bux6%C;}13n+P znbQy>*0+&1Rc--6oxgtcn1TlBUcR(@b?mP}5?B>N_zT={7jc42C@t+oeibJcoJyg5V68a(nAW0|l%xkTX7xNCU3G z>{kY##$m(Yr;$5%x^BV7VQb?$JACK%OFLhqofkhq*sHWEmycNH1$Wi2z`VFF zO|6>s1^wM1N)xrskh;b{2=74+8*Pgn6FK@HSwM3r$F7Ge|7*#MpKAe+XP_EURDTdyqm<##OXvg9*B1+1i=Hp)2uwpK4MuY3_<_=~@GHq6Kwf7SVldwbBzuEbe(Q`dISU9TX;zBUqur0sO~MdX?hf%$HhxCs*HO z=F_h1Q)Qi&Z|@Ls=iVB-4sha}z zekNux@cm_$x68`n(Pf?W%P&j&1TzhQ`QgJ!5uk+HssBnuS(omn`_>8NV6(j23>TcA+G7(dY3j)*7JVR;81|O7!zb zf4a)@$rPM>RPWUa!6hC$`BE*`nSj?=YWwz&8Lv-5kyii?f6 zzk8Jp*_pyFv$C?j&-oBDGjlm=`Kru(zB`qnkn>n8bI+1^>$nS zT+FJ$r9T?P^q;BJ6L?;O2QK;~4)YRpNsYna?DF6L41*^%W0?5xz{VPhChoqYJi$ho z(pzswAt`h=1VgDSb1)oXVt+V!|Gsxj$3b#<0U6j0qIi!-s)XV3e7&32HvGtki$;_I zC!&tb)}w8;X}4PH3w7~S?0yK83w1VbVCF^F?7gHvmuOAk?SRkqzKEut&?5>{Pj+%{ zqfq;&Nj1CMxReuTT^3kiyzbkr>IenBM!(AI^g*E3p`3$5l+6!u-&$mV5fK5{#EHwe zf%eUeNi;|pjj~BvUcVz{6NHT9rP%gc<;*+P$gns$R;rCkTIR8#7yayBML5Jbp4vW2 zKN{M{9Y`HW=j7ZU7WRV~SY-->Ot-%1riXYe+`tR&?1_LW$wAfc#?q#`R}zUZ*ykxp zM3}AG&oPUK?=W9^=^~{gZY&z)ZAeYlQ9W|1Qy2DvYXvDa34ZB*xq@{oQH>`nmjTuy zQ7rCDOKneqp=@`E?cSuw5blyUUW_YTUe4oq!ID!{zyAAlc9a z`S2zUpGF#0Mlpe&NHkX_=&^o44^S+%Q zo668uUsrKD8rK!HNw8)h`Ha2~RFAv|$08o647*QWd%^U(w4$O53}vB!(ZKjYhm!{by#Z&4-XH0kzX7z;{z?Kc&=IKQID}&V(`o1|{DMH*$DAIC=i4tv zF2hKxC^o)rUTbZlLL}nx7v0<#wGymr zjHVlAe;aJ_KOM*-deJ8deH5-mRN2{^uUs0hZ*BFB`hz!flJbr*zP+EoJAVV!Hk7kJ zAg!ebmsgVYyOPU3P;iGz9DJhz&cbBB>JZbl{lbzHjUX5)sA(sDM8CxQU9l8gw-y1p z>>l1@!iqm!8xZFH_M@{CGo<5PNiHKqqYH?WWA(l?BZ)QM{|w1k#6hDCdE0#iD5l2> zoje^k9LSC^U8-=6!&*$l8`mi{^S)N!lRAL+T+w|onMVs^U7EVS1$S5_FEP2x<3Ile zcSzff+hN<0Lv}OKxw%Jub}K~w4#-GzX~X>vE-@;bcNTPtl^^V38=B4%3-9~H8yx4~ z18)bRgtucdW*iWtkbL-{XBBCp;cL$DTGwhKQV^)&Q_b(H?yq=#KXis{Or_@)U6WjU zfdU(b+bzY~xVzS!zjfn@jv$>SX=XOy+~(n&qyXn+BGNjKZ#Zr@fA-T%ZTz{3Z*-Uo zk3O<5hx>5E*Jq^?d_W6lA8;qEl`qA6X{A0-%f)8=-YTU>%}|M1%A#;agdvrOvetDa z;GV(EXO${jbRu|iQUG}kyt-5`O}*lDBZWYE5GJvx=N5z>w)NJa&wVF)7VE>fkgHc;ABtsM_n^2oh;H-Lv1QtYMbw166#TEkJyWbU=jAS<-mk&9&0^Vut0@?R%j=iay4)C+PDqy! zK^9X}=+7=!##oqSMwo~Fj>f@w6Xm!n@Slv&b`ixhn5Lnax@6~YYEE2L(5-Q4Wp4J5 zkh<}l96H9q2NUG8q*Xe>UU|RkeRC3Mao9*}`{wygPKY~RX2HzN2@j@?%zC(eJsZv6 zS9?}S^FDAeP_lsMC2S(X+}GK9gpK7-PWIvb(=Hl}xv@)LADn8Rt4p1#x_D zeAW|X-izdVckFzwLb!RJV@>&u-+6lXuUkgo+LB5Js#smPTFtDYqP!6W@FVP~S0_23 zkL=|tK%9~z(O*{#zFuNwaKE}jDhXbSteaEf3Isf*vTak{nD~ohe3~a`4P{H-lJg*R zpjD@Ly{$qfv;8HMUadfn5}>U-v2OHEiegK_j*_nRIuHxwF49K>kAAmhzSUaMN_dL( zwj@mji?G>2)RnKu28SodpQ?YJWR*o~oH@k zFmi(BXJ+=XY@3PR|F6hgt1k%$hk<;joVVg|;ao|ruBf#~D2pQos|NPoG&OC$oYGS5W;SGaGrRj0#YdT` z;VQ0Og}8i;KGONy#D&`Q&_6T~fD(0dE_@x2P6>gT6zdmwr&e8FW;}JMZGBgAp49c~ zRFs4fd~i<43yT8v3-Ik2dkx>tvALFyd<5v58Oxq`VgKg}o=@ZdP9rf@4zW z{)3FNN@M6YDqLGs%-=}-vD>CEh~aBT>Vp?hrM9hnp?odjTGhLpq|gC+KP8KQAeZSO z@O_*)heqm)6tX}<6KCS6%X@~u5GH7xaDpt#433!Zv z1>Zc*_4%D}T-_CTdtK;*;?L~yQxi`09?3q;}*VA5|q8k2AV5kxskOi9OtKt`1Dp%u)| zsA;C&=wJmA5cpV1#yW5Q@05 zq!3I;8gE2j$XS~KP*H)XLB>DD{$fseHaZEC1)dmVS+3G7qCZ@dy)kNU{I=h{R&|#` zfUP>$&*)d2tVu=1DAabF3cGlgoe5h@|IyDQV5={ z|B!C01Y3~+5F(8K_5uc%8XvKYx9PcCpc&8VaM={4Gewi{ef;;q2mzGW@VtOChip+T zUiab8pO3U}z58Pg^JR+_0ew?Z`bFX$C_h_W_UniP1*$QaLZs0a*Js=)WvZL~ojrLs z>s1Oyt}M`E*7RIM`dQAZR;J~bJ89TkQzhf1hUlvq1n5g5U$$#kz^oV{$@IPQ6bsZK zp!WWoU12knzWW=w60O#L{DH>xr_z#)=ha9PMWd=9)S>F&{WzVrTmq za+!p9($A)_C-?0xr$vf@WZKoerx9`(H1O;15n5Ba$iCBfdR?Bn$W0dSV7hc$N=O3u zOjm1s)kU@>%Mw>S;z`T9jwJBe#h&6L(lp?^wYw5Zr_MxYa%5*pB2drcMLpI5)as-5 zv#);77?0MI6Jl6T!D>A=vxVuOAFi;@6E?W4u!u#x8qbXbTLvXV(2aDlM&9ELUOWg1 zBglN?c37F5j0U_0G8k$zf}b{y({_QIeGj;Va+!?*vNQTDRU2o4l>&agH_shB2yH$Y zfA!I1Ip2~)s>ZmfAn7c-DD_*DM}UjCtAet~J3 z^>>WaBixcFTpxq=Sl04q^V?@E-6Mpm9*P<@C9-##0s-G2sXiXyV!*@k|MQ^(YYun? z{#m_*d9VkFlEKSl+x0E72&h#{&PN0_RQd^b(m}K`%$JZ)tfVQS!dPVW$4hhKEaFa3 zE6MkJ60IoKbvS_5P-jl+TL6psn(#!WB3++rpI>({Dwrm=FuaWuH6shLS^ z9os&;xKOO0`daZL8!jn9myAf@LuSp2pSv0wb-~W0? zC1TNkKfdhue8A0gd!YrD7@2P~yEofotgx5KPMesF%aGo0?@fYKu0+tm)q#Jhmy>a6m6Zq$VsozVy2v6fNEyUr&`2&E8FM)Q!Q?xR{Ew;lf8B%_!3oK@V zzWE$;oixgBYzSQP5uG;}A6Sq^8=Mv$wkT!J(6}h(W)OJ&}?>+t#2`ANX4mu-J$pau!iDMmnL z!0|i|ro2lSbhT|xrh9$5{YCTVXFVEI`UtRb@|r5c9zjSV=j%FB{|r-nvHhS8lkz(o ztiV*`LiV;o3l^j=XLM=5jIV$F!)PS>z$USREDPF-Xguhcw@m$lRLGxBBX+=dZfPS(}N8Y<|U15oG?OUu9OT zCk4b0K`CQ7E%pMBgKzpM7~V+wAP@uyh1=&f!L-lUeW;g!Ew|jXMgQt_`^oLqzo|X- z>l&O|Da?Es6mob$P<-wHsE~qK*Zl2^ruFkNXfXL%{xw05Y^)4)95?aN+vE+f3az~v z`gIjnWahY5+F*cGhfgD%)VOjNfwB+&?|N0>)O|VqD0nVH`^E2BAX%$WcGFb`!Duyd zu511AeW4Abor{&G5@c|V9Pv+q#)=V@S;mp+FR$4;eFC&HwLWR-u&QM;g3cbY1@kI% zjRN>#{X$<1Z@iw+tZu&mIXN+MjhLnfCE356w|DBW^MjIWaqm4P19DsrQ_bf1VmhGf zNCK6~ZzyC`R{Hw2-MTZpHpd;gf4#cy$KB zYX-B(>*_e(r45cJHmlGvIoY7stnC0!TOZi8{7 zJ92Pq7In{|1Un*8^`%vMF;!b~BP}bRfoR9lkqnaN&{$?|8{5B z@9$QErtBE@BGJ3Q$g*M>0 zQzm+_u)YY2-OeJJL#fpR?No7l`L?9UxCmQPpO(dv=O62l#m;{SMZp3SY<#BHJ%{)& zw}#5WuhcaDD8YTn9DLK+e;LXQ^I5;R0Vps4Je`b>u5K}2S}5l8|* zD3Ad6aAoo$ma3Xg&@MZcR_ytQ5uFnC&sW!?kOIi7H)6GYF`L7nZ)-Z}1#2tV3O2p7 z1aSlwkqVFYl7ipmFVtAGa$DPx+N9_ZkC#3R$#f)zp(1F6Z`jO(Zkt&4Z+}w+RYBq? zC?U@(@PAmX>cf$39O=QWcnX7A{YNBJok9jv6`yUb$u4$Iyj1z7u&*109CW%J@bMr~ zX^8MM2s@_e*FtMpk=-^%8mFy7BL1B18y^SVjd#2zrk_c(?0Nn=-x>lha~oUZe5HD| z(X%<5hTc@EqBQ#%ryYSyeW+#hv($$aq@rzBcoD>xRDN+ENIGL{Y(d*{4V%ewxtG6eo%@ptCQ6AfkW zjhP#G`m3Z<+870TNFDu-XtWps#zk5gvKmzjkk6M!^llf5aFn2~qWT`T-NQ$-f|AVI zV8fhMqB$4G3m#d+3ptLPK7~{SuLb!r-~yD#?ZYlkG;BWeh8tt%m#yuc<#q_v=qh%d+^!785aTTaqE=9xOFI zY4LV?DU`nPcr^3t^vly9U#8;4)o<%Pf1zS%$(5*kAyFxiwfu=Dd$Hbf*sk{my#bFp z^zRVEpERGuxrFo7XX0lfwnLn+i#k;-kbGCWeLq|1$i+>%mzEZP6(mR7xz~2a=ftA( zxxYKSmwkh5E%>o(V5te|SpO?OsV-)Q7kd{57FfmI8tcffu>P@}#itYai;9a4>H)7B z39(F2rh}y;l#esBMT2A8!k#}y5?EA}-RQaS{RUvmD97biS!?B&a`EW6=-6?^BQ$)Y zOTdMqL`=e2mY3B}FZ5<+EW~(#(&@7yC~I~(i`W@Ycenz{5qU!)sMqU7Y5l_^87@kr znL8z=Q^QkbXO_6%G%>tL zw42CppU!d8{+nveJy((q#*7n};={J()4y6S4D^_3Xj-BQ=* zIo9R)Y~R59JJDpP)#h16nWo$!0XkqwAf|H_{4iYPUE z?-#%3aqHzBkq*3aD3Ig$iy(l~Q=~x@6EejSpO6J`#Gu$qE(@mx98+_`rJ55J#su76 zuZ};nonW+T2W1?1NSg3Hw-Bj%lt%1)xUf0WJj+$D>pRCzx?f+fy0l_6$@HQTTcv+vRv#Ol-qbrn;irm zf6WhhO>d`}QDgNNdlQrt(C##*oAQ)n89*O#ns~2jzYj)V*T$;-sW3;xh6@$G_B^Cf z8v9!~*wcyHlI zKKCt*bpq)NUSMLYl;hEZXtE&-T2mpVh+(8?1jkxK9QHvyb!K6s6HT04CfnR62x{US zky2(sH`-vJZV}SZg`0mfLH%P)WqTyvIPOw3WTaa2K#gs|9U|0ht}rJQ^ycSr)Sv&> z1tcMbxShpJ#>Am2p-&*M|tGa*O_yfF8FKzzqe1I)I-TOGhk0Y>~ zkE8P&aK-_Y8qE?ZJIq_96ACdH25-w-l8Akn4D0h!Y0>A?nDqRAFpZJ$ z8Y2H+Y@LNylyCUvhgLw52B}d%y1PN7yHh|)8l<}vkZzFf8bZ2D=|;LcB&9p{;rlzg zXV2OF2WIAd=ZQN$*LBBCAv6e1njMp-7_rrSukZpsnG%(r*I057*_{N^Ot4&R4}}O@ z-jUA(6`AaY*qLk#hl_mdH0Zs#3>;2gRaFS6Mu1yTQefhIfum3LeC&7teqOcf@6w`@ znN$f|@I8dp7u7em9{CD>H<3wyOnnAop{?DhO`~0BQ;N#r{0t#Wc;k3o0>_soT6~9# zmJc%HIM$&r8;eaVefIZ{MEqx7gfW|>iV7wOX*_-?yzmQZHoR2WodriJvs%B8~0QD~l zc|+?4mw)NB)9>`|f-57>vn3*zEIwzQNXmxkzq#)WUWZj1Rjpjj3hjRRiX6^yjhVrX z!Kg0s@;rs+_a=AWpqTim+zOiedlr3vUH0{UznESGYJ_3_bl=H=ZY>VNE$i z_3y4A_hYg@Z%vL;#=qsZ;vV?F$`$bOL@6kvI(gEYJ!^jc3Q_{JTDooCAC*XNMw>wRO*!@&jw zN)0fijl$G~ZFfbysEoax$hohPc*2|=OGCsSmc`hhO>YKM5f6{I$eNNcT!Z=UY0XCjrMaqfuOa8Vh^ zt^r6C#G};bnOpni&~k>q%O{aP%T3;E)m2Au2&D$H*82N$rXB}YM&4~J7j8@C8li1- zzyCM|Z(kTqYrvx?c6UAEGj%Zz@RnE*t6G>x|Ar(@GZ$<8OmI8hGH*x_0!#qvL{_>m zCOQZpYJ7Ef-_-GWii1mYUWMW{`71eKyLMb*REbGl{Is0OPY^@N5UL?3KKIi;MYQ~% zej^5KfR8+fY?urr^nn$T2RfuDU^s%3mmCOyejqj#h6`hYbf@9FUsZ9u+0$)o=6cX= zjTk}J!k0318C#r-_y$(z6Q+@`++977K&x1U;bn;W;_0VmvH9Ks#M$D7n0J|OLpCTC z{7|;dZ+Ctef*%eJ4j#)CL8#Gi+!+&_7sIRdBLNME=T{dQI!E-!2N7ABR4m<2j`Pf4dmMnq5P`d zE}2o6D=^w;4#bTQ)fSHEH+L&TF@d{oNPj34a7F7?;^S~O7;yvtRL~cA0w$JB*%h-p z5jE7CnsuSWrIsdGXMkIX+@5ITZ`BVv0eo@bZ|1Jp&(6aFkC=yL+7!$D^m7J%XS=Ks zrD?!BB}YHYlS!=hB-&go*MzkCvhA{FhSmJ8j*GC<*tBp0 zNE`NCCeH06Qs5`xkAKKvRKX|I%$G(wD3EFgD>r?jR7c~!UzjbP31aGEpI&OHBpYkM zrDBFS7@&6qPx@spSbaYdX(L3SZS(vuSZ~uwRqJJcpZ>>R=qW33Efd^=?q`Lm0v_s* zFDa&rR3wlwNaQI#i63MFk?RUfK9&U(*oD*Fy}HQSG0QiA`xn(fU3&M4d6{)`Kbe0tI9O| zHw)=zlgmG8pkUYp1J!^tb-r0%D5#J1g4w=YauX?>*2(IX-^0Thuzvgq_Go1N;bMpE zz3@u)m2gAHsvf7VYbAJqU^VpOtqh&55kG^^>n{oTG7LtqBmIY?4GE+ttbW7WPsj$4 ze1FP44t5O4U#0!`(9J6xt!}6J?{S8wHlsliTl11l?ahswO=~y zzljFJCK;d-la4dLHo%crDS&H*n9cUIR({~O$8^5|HM>8IZo*8?s3ka#GooUoLf*|F zY!VeY1B0$lfKnJ16Lt>o)KHjq=6ehhw_cEdJNh8O(Z|Ykg~_M!_6;VnIqWf}h??HD z-~9TDVzND)hI-`7DlFGxI+eqV^87mm5m@>w#xz}ONiHqrZiUfS$33oHdm{<+gms(B z{bHbEph$a-ohM=_TD1F=FORf6x5LMfAiO_i^iT;L!6y6XPbbfFq$!&z zap-Tb-y@*Dg>}oGr{MtQ`0$#`h8|4iif9JTa=T*bKuZPB!jK1N95)N7vjK2iPKJO8 zo(sgAKBe9p+3nBlZUI-(3p6|h-rK-&hPD@e?GVENc;BNW33@d?EU8Wy+yml*W@Y^0 zraPPnb?^D}=P=pvKgfz4`zLlH1 zU6xy?uf2>7{?PC2&?gS6AE?x+zaqbg(a`r&!#;gU%nb#FkeZw5Mc=UOuu2dk8}7_h zZQ8A5{Nlb=F+^P%^}2KwX4!$#t$2 zSK;YOc39@!fwb`~dobVxNVLuUmkO0~CM;zc zj1hVZD{7#FPrH5poO)`ueF1eR^zk$>^% z`XQ;VcB(HI=9s3HfAV-6+qiN^SAA73P$t0-n*9>{sjG zFX8O@eSZVcDbyq8;AOHLV&yz(wHnRWM;rqVk8M7W=?hIR`boYf=61*v;wSIgtupFH z>TDL{grPDqv2>PUD;TH2#$K&QuO4_@`*R(OSXk;&Rkp6vI2}2V& z^APK3&8eD`AHW==UO&vmL@Rw^KmkiK>2%13qw-lC60}L=4bW)%%v<55?O{v=-T3jNMzcMq%T(5t-Ac4p9UA#K-8O#T#HEy zh5d!n5`6Z`f?4i#qBF{@tMzhFso%U*PYG;nV5=zwfz?WnQdP#%mnG%C+DG_urq({57QBmzZ8fQb;0Oe^b9N%~V}7fs?zH^;&enkz4A??6gZaR`r?ez6owaF@jv^1Dkl9tLvwWeWtVUYfz{)nJ7%m zwOMqJgtOaUy$}O0>y@&6px5n15w4$51@}?M_!e-lKJO&qvCsXz*B47Q86xtY z_?6t69)(j>xW6pYi^dC3_mj;cutRpd_Q(+rpU?<Alf0Qu#Cp+5>1Ro60eJ|En2Zdov?^ zKlY#)&Y-OD+go$9Q{|`8_~0MeX(S!Ab){Q%J>1)Ex2mK#EN?|I+gD1lsG>5)+Jf5W znSBm7UmCd7feb(9bnlMgCI(bIY!D%L1ANJ*aLs&jX~Vc}WOLZ?LW zA(g$;=K1TrKStKa?rBUW>+?fC!8jWSNUpYvjd{B)Mton^>TOWEoM*~0Sxx%ci`0s3 zYz{F8^cx*TaF;-fks3!KiB%N5BN{0wP@E#qlyEsON@>sVMu zV*e}QVwS-mK6FSDD&7pAYnQ)(KU!*R4!&(Ds_Z>$aq+BroM9rRJ2wLAfy0{UD*-?l!x{T~SZtB_8 zyZPAn%v3UrNu0JvJ(^Q(UNW>3pF)XN##CImrSFgX*HHHNQ-+x4hzB>6n123ngtko2 z?Yr6}s)TBYuK2OY_TpU|7?wP4yJGG2__c}*l7Ih+C6k~T#``IW9!#;LZkFx*SK=0SF~QHYR!vnt-KX$_;&;1vXx& zA{^mf;4#AbsaQEYs??T@eA(L?i^;g$-2UCuEd;yD!5T}A&8;CKTII>&X<<4^o;H`( z$m`f!e!P&gGxi$!a#=&e7!d#Q;YTUeHH?stO_u8DzmxruOKYR6+ng?=DFL!VO(?XA zq0WC$@Yrm*!9nIS{Bz2&Jw3%~l~0`WT1O>uPZC>!$>W0nOx|eOYPmH>TcbR?T=xBY zO58)pm6?FTiO|u}vUi(8{3W@>%f9dbkWmwQD0Yav`vmz`5H_j_p&Gt0Wzh@jt>r7e$#=fWIQplwv^Zj%kBT##(nr7( z16NaOqC=t6^@!j!`RiP_FDvY)QVAi#`W=9BYX*siq#q9dJ;*cB#=2Qth&hFVqOS0y z-9FLJmW$*BFC(Fmn^!4uap+QVL%lRXD)7r^Bd+a*0+77M6<2FYE4djvnWeNn>^Y+7~?DUhWjY zRVE1?9;=q|x}QZ6^ErRG#m>#kqg6douTw9b8*BAKxi9*d5~BE06Mh0z_F_N=W`1|{ z^75b1Q3awfj%^8)PJ>a!hCt9W&Y$YiD{c?qV;AXPBrINL7|uIdKG!kDn)pPGpHvfxkIzqC`t( z)^7&HD4Ial^rk?b1mB9@m&JlV?03afk&1?%=a(?YUGN^j?ZUkU*9Bos)le0V2tJl@ zzrkzto`A0;WWh&Pu$iHUjDtQm8WlSgAr(YXI?WgwEB%Un^_f_Jm-dH-892=gzXsB0 zT&T=;!{f|S&V_Bd3luY}0j9*%z17X8sqCyQzNwGKCbv%#Jpg|i6@+~Rm;8HV;}vcW z4i&Y7Y<|SJ8A7#HuGB=oQZBj-2}2{v7OB|Vq)Q4mzK@2@70R>vVulp*@LJ6m_UGwS zI=Q!PbmHw{ml%kEz2@Cfyv?Rm1GOrIDk;yZrCFx_2QTI08t;}ec!_R^1>>;m z!)IW&Iruq)Kp)gx^&-fnKXH*vWSX0MfZ$ zxALgRnmp4#A4TAjM#?{tZ$8{NRl1!jpk}d|Hox<_Ia6ENawXWFt0_L!qIZ`xG^|9m zVn$*HyUcQ9<=6d<@GtkmQBML43)a>(Q0}7phR{kq3*k57_rCjKQy?Cmj5^8>eN`}5 zv)B6_lSvvlvDM6pSNr@Ew5h0v24RpXAIGx#MEL8vCJ9Z2M@D*i<8IvPgY8U(Cq!M> zY!A_PBZ0$l`1PhQ@vMh>K7WBYgav=`$yoqt5v=H9fAkNa>aNXzclhqDO(>Gd{@
28S;~r`@&Z*|%K=(VQZy942UO zhl$jx^mqjC;Dd(0xV6l>tAqrvc7L#RS_1PT%FG--rbqPfFb$ofS|y#BQ(9I)zR`7= z%IJ2RA1}?)&F<87$q>0qy!Up0R)hKy;WD&yGG>3bg<>)^WdF~MrGrvkz{ z#e0xj97Y4(nR(3XS9OM6SV~*-vfZq0L?mHasNT!3!&c2WhO<-!Eeu1CVvaBk>bUGfVUSkHEAc{ z7P`OQbzoRJ02BMB%ieQn4)MBzUaMz?KLakKcCA%j`GC3K`BSKFzdxfyvOH6PovYvG zhdPbVl%9Bhd)c=;NvBHqZ>A#j9XkBUOXhAF5QU%l7al*3;!Aw-Ng@bW`J^dWoP;2Zt0%=RmVB0k7b+{;?Gg>XMIHd)(PW4Mb-ic#2QX|=9gi`U4yZ-ckk0Yj zu4=T&2!dkyU6eqGOldh?rd0o0&@&g-3Ai!UF8odBiNZfO2&1P0fvDhOD(VywdwIe3 zd6+BN^?pjnR%RsKtELin^8kQbG}my17_7H5Egp#gzt>mWAQEsdK__I@ZJEu_XMGRN z zJ#M=jXfvx!Glz$#jupNFzq6=vzZtoYZu%?Iy3hU3$Gi|Wb@T+IYsH9^1B0LKrqOM+ zhTJ_Ci8YwiWzT+Ck~1>?oUGnS#qH<$sJmW1Ux8^S(Vm)^AL575PYLX>)Kpvaq-0+O z`#36Ybem#6|4vmPB2JCUTI7Cn+!;JU!btPb4TFvgQzVyBIW$8^g~6PLDwFa2^D%H~ zx!oSP!bswLD0QvwXGtrf5*9=Q#p3>@$71>^&P@b$u?oDt8_ayp=iUFJL3}B$LXJO4O z@RIu-WVS++e$-kv;~{yDlmK5Hi&hsh2TU?zf#TK@8v|#l2d<$2He1*M+;+rU@0}Ca zZ5C9E4Y;8{G{}=&v%lNUp2ULTf66e8pEN23%JIc1>70-gFt#Jlm)~w>1*BzOHXp{_ zlt=4gh@wuZc519|_bCKQJV)eWNEQ6}8Djn<=~KSkq+5Pb+s(CKsI}Vn&*s_GKPnAN zgHyilkVyojBvc}~WcKJM{rc3HJ#|NQvp5*%i}KtvE_*$+z2qN`_%iuZV`V~LZpc(y zK8HQ%XtWS6Ck`FoLy(w=_w9y;B2hayHZh03&k4pl%&Vo54@5viBqZ4Vm{Zy*b0@ zp<<2%^7*dsDyWEcmGN|{#X~`^mCxOG2N$7OaCQFsIFZz| zw0Lf)l@eHMm6U>uu2Htn<^GD+Xl}Sity{`CpQp#H718Y|ZHaJR>iEWB8B;`C&sB@S z)JxySz8i{qxJRMRmQu4f7D}Z(i*aaq@b+znoOT z2|2Nx_{gYQ$?Fzk^3mc=EPGEE=Ak7rqq-HACbYhM7Wt=9y=MJ0w;B;6ESTB~o6S;l z1FRF=eA5M=d|xx`)!SvgTLIV_@8(MV3ePckykg&>T52Q-RiCCF3Jv07RIJ1oF}Zqd#byOPa5@11h-fM&=uxs%}iB-XdU0k&^d?-R37D zOn_LNvzRvoOBVjM1^oa3N=2>F-O|oZAlKsxkFd5L%Xt8-1sJzxoVxU}UC4fWhjlFJ z&Gwd%v1*+2=ZgWAiK~cikVIYvr(J*74AwK6BL?)v_eAzV@<| zAz-Gb<{se_31Lm{`dumN1r%riWLl=;oDd4(%nZjfMaW4qax{g~oBF&+bY4wzAMs-| zrI$gB*G0S&y0Y1a{Uk4r*+T zqX4WM#Ndsg9>DbxwcHB39m%Re6Jp9pYr(9%6wU-cvCd-DExFj}YTqHOvABEOsK^^Jr%)j!l-Uf*FG<^74wg-l@0_9>V2b=zFDoS|$^wn2 z|56n#&BdhSdV$i?k9Gm4`B&cLXCq2iHIc!$e%gJLj zl8DP{MpCBfe0P#bLdibK2!B}< z)SkL#y_C>sYRxw9oCQ$gbnWc5 z0SkXJ+;DIYi}&xJiLXMl$fLYp);^Td4eq7cDG;>dBDsM^zx~zG>cSR9em53s2y(|^?2UDXa6FCM+ioNu0a!P? zyqZdl!O?|$CGx(nik|Zo?KtDJ-(209oyJUom1G` zZkHo2XK!yqOgY7)zOAqZ{{j3?xfJIW6=ZrP4Kxc5ITQ5$rV20V>~O-P%qJhO=5v&(9d+E3z@GW5hUhU94g72*6BKoc%%6bWR1#d} zF0#kJEtClRm9ku8yR4TIa3pGNnDaM~1%2@)vrHw4&|O-nZ)_}v1v5Ss<;TCS2>Z!Q zd@eO;5kZxVj*qJ~5j*f^7F-CES21#YU3LZow7si?W~&2RzbbOjZ|_JM1a6?`UA_7f8>~6;_}$3 z@MA;3~uUv0*-+ujI4BliR#*Uyjz*c}i>racJ|XoP^K))_hY=sfY`YvFnfIpf;Pwzp(O49v4lz z9j+Ys2t+T1-kuXtbOc2zMsei*kzJ=eT=ZqW{J4vnVu5qVHxG$8W>rxS_kwIuoHKm9 zY59Fn{wc?Jq5e5)s&GvXba|>G+Xf%8HxPeo=vJNY@E2-D#c@pZYIRBXS6Nx=(?IuD zqaM!dwNQm>G^oi^;WRo*l;hupNa9g3YXvE{LLZds3bS&i$ZyXD@Ypb5SAJ*I)cr8|JB824i+W4W zR$>a@Rp(x!bZz&&(%-T^Hr2(q-TH_U^-0gR8PO0X!(3a7Xati~XBKBwjHoena4hQC znkgqI<*nBLl-S9BR~IMJ0sm?L+qGBCyFVnj@%EoLwnlGwcp56_s1>99q@hH<8#CtV zBJt=YFF~tE4T9vYh4PA$-4UE?B&09`a3EDbUCF4vQ&KX<=dM^ za1E}ebJV?>bd{cj_MDr!`zKEL73zSU9DmP0<34uZ*o3a;a`U zq)#{dP>YM!8{IWh_L22Js93`A!unyNE%Cv9% zjav(qp8PPfA6?hud&iaRq!I&uMx!=ql^(=yd<^Q$E-kQA7`mn&Jh z94_SZNP;?Vp~172#!sp8?hqc6HN|A5a6fgd3rTZVxcJ;kNwcj_bY_l+eTPONksg^?C`TqE_MD) z4~G4-N`CO60d8jPrv58IEePBaxbG4?eGMJxHANa7cW4jhoD?ytk{+MbCTa+jq~=HGw*(&qEo2I(-%a?uk#^&M~D#{GQjfx80`56xf@wK4R54ib1Y8= zHFXBgj4b@^uHe!g2KpTrL%|M;Z-D>QFa=C0k(-iI4KOg-f-83>3bXGZp$uzVLwl${ z!+xF7OG?8^T1~BxRSVlqg%g`WiFxZP#SX2p~M^c7?@K2rP^juyvMeG68GPvYO_uiVEp)#4D|Ajq`&e0 z7G?<-RSf6hkpxE8B!pRa_vQvx@V4>`%z;XbK!&KxKOR>A4b8n)-dsuZvfnuT)DU|Z zhjzAdIe-TjRotx>*iIZu0#*Z4ExNu+eT^$XtqI5pOpu8YXe46pQyhz5(g-%4`Z$(1 zBdjAb_|a0edcv55np2FvP@N~i)QCf?1ka<>P}Pv)6WK~uYdHx_xY`ywL7HbB(M1!q zO|hHg`kKhVyEud=|BjYLIK#)Y%R>xzc?v*&m@>^A&78lCWP>@}BTiR?Bke7~S{_4e zR+O%5P-r*vym?7ZkHlyJ$20bfcQ1uWjlWk!KVMgpkLuRuAQ~-_ks8ON_xA-iX}h>o zc&=#8%du)@wBUg<3auo0Ja;kvwh*k3XuV>=$sY@H$zO&068E!6++K$0tU8m8RCN~l z$8GwrDUDvB&3}G4e_?<}5rxa%`k6#VBr}~%6Kzi80foFuzOZLiPw4R>-P!wF)4k$w zPmE?aU?`Quc=Ui>-hYg%J)>)gQmA?7Cmf^oVz;?inx@VWKZRNTv;dYc`-;M+-H;D@ z!S0>gr;>p}Y8MMj3=%eEQxq-2&cd;0g~oqs-T8JDYjS|lPVw;z^UBp;_Pw||f3*9B zI$((S#GQl3B8A!E?03jo7IVUd&z=xb0*9}&o=6~t zQsH?bdEzk}Mf6yiCPg`axwGPeF6TbR{KFvPE(RkuPF{N0Q&GF?} z?81&9_N49KYTsbn%?aha)tE@`#fuWN2 z^%d7{-*v^oJt%Gg(!Y}c`h(-a#VZ}PzPeds8B7101^7iZL#o#h4pbD+3Wlls?t*_? zS-eX2Dd6UDLb?8t{8c#Yn(JYawii?Ww?l1XgBhA2K!empyoVXT+k4mJMGagd&aG<9 zJaK;p7K@qh3kVv=N7JKUdH`+M1R{UqPL0{0_qVa*{x<+z)Fv$h4WV73vO+0Ub&-gb z&dTBhFwWIq_wA<|qHyN_P}$6ni{hEzo)BmM_T&{M?4*qrKvef-i3U|CJ&RfZngB4B zL&JD~redz=4h*T$mS_Mc;#ixH52!e0(f0jWifpESQ|YT4S2*h=%?dtKhPBaoUhL@+ zv&DzYK9eg}{zs^KR>($G*I7<8mPX1BnD@6SkrQ$>fO7W}P{XP8zIRvq1}L+#DB#*;T1v}o zK1P9L+}sDu*W)giee?m;M+N-7697*%BmM@nbGgid{FVIEGQ(zQvNC_dD$T{_KuHqX z3|{X`CcQ|8aCh1DDBxpG3sFenlzYNAkksr!rz}B-7s_CFz*S-&U0x3ga^;I z+Mgc5;^`6oCuW2uL#}sl?MWl0#c-UZ9hPJq^bg&fMsz+s>LnY6Zh2{eRpTt#?D4PV z(im7U1Fiitq)%8!k`tmdt-$1n5t_ri4CoP{)6Y!^16AjKvWFuSfTsOF?+U67nSCL1 z*TM{8{Of|3AhTyhGv__Hu$94g=`+vc$yy7xv~%l@Hg)e)6Ltf1H==-+vMApF#2PtD zDXKz~0OJIt(ogmR@!+eEqI52}@k;)vPLqlPqp2jtl$GKJR% zHb&mT+}s2cen-r{=xrY2B%?J;T4xwucWzH<`TH1fl!b3qI1pIgV5r>*R)*gPkq0oI z$mDpf3$!J4u0b)LGW<1%vT(QIm-X%!peGaN4?@NY{=RGT^zOb;JxaB#srmtC)vgeDze ztDIt^^RloXtLi$}xDDwnu`8o(bo0l2sb|vt9MOV>j%+{j&yEos1FIkCmMi0IfHh?} zU_>NH1>AF?vN5E6Za|({K>&zl^+U84k=iSD*(TVx* zz~NptAA1LvYDulVR>RbvfR9C0psa1lCf%Y6}Rh6 znz2d~nd4idv#YmmqgjMcWWvTo5w<c94To=$ zxrS6|MP+h1{#9IjYqIRS20F95k>*(TTZ4oFe*qz3H5CogN_)%ywD;`rbGd(#H7$*Og7AwvEzy}y-)_)#Z0vkURe7Mfy zH$pUKdHJby0X~z%g;}7+mjl}(705v3roWahls}ZLJ4Mbbl~!&34mdrzzdYUe`=^7| zY?PT$z{B9;4Ys*6`IjO6-;NeIfbW@=zs|Pz;J3hvHn?3 znUF;9LsJc^{fE^mxemfQp92vj^EgoeZ+=&*4cGgi z8?eyQ>h_oA6VlBGV!jBS&CP$*wIGFy2AHLW8bPSo?FEU+S|<&waqlxJ*^;rf`(2&&@$YJtHX60^f7~>LJd1tyUtzN#gVRPET| zLB07n@@)OFM1%V|40JO;G`jqw0VXsFv}!O|xIcz3`rpN7H_`X zSB-?bE>3Ub(ohg1Z}b)+JIHb_cj0q8cT3bqZov*V;>>%5O+kLDC?_Wq9w=d7QGM?g zuR`7Ip5sO>My0crmNj^J>LbgO@w-C7hKYWvves}gFsK>H>~}|IP=f}S$}6t zMT)16AK;pF^5xUYd{a|(!Nl^;K@y{gfm!|Zj9J${x#nx^y^$m~DkpkSK1o!{5g*8v zQgPxR#GrC|2to(Z=NWDvB77&oGT0(-@fmeTr%_z(x!=UZ#;Qr#OqZpx+b-!o@@c16 z%1~3dV`^g9ONmX})KP=5;eP@u-Y36~hVspG$r4wJ^HC?B`?^K@f9)L4`TJF$F{0ZK zHL|F!<+}e29J|98=$wXXRq_whotaf=5d<)kf7Q)Wt90DCq}7A7XukNj&`_*eY>Ly> zthMDm+;aOnz+rBIj)g@DB+Rh4i98no6kNoEf+p$v^5+B(%N1NUgah3If+G`vYOtmO`(Nn2$B#p0_ z)(tdc@&Gry+v4>aT4kHV?htkq#@0%A)878|JFA{l9!vG zA3UR#&Ig1TqI8OdC092iaJi3fUxz-kfi$&ej0X>$_9l zI>Q!jR!dT^T@#HLcfPU*TP=Pc;s~A58^x>Z@%$k*FWc#|mUk?#k)~htGGiPC_oQ^TgSfDNqUfjOeC@&Hu;J^1O^v8i z1lHOKz$nn}1Ig3}o5e;3KjD1Z-cBUu!_lm&?qU!b11&`XbV%2Bu39hlOcL3+tXRD? ze=n1<&g$A9XjMOee<&XV&(sFB((aYZo=C**&(RYuZjgc64%6Oz2N~!YA!A05#vHJR zNz@p6$n4FzpWYa2?yxuN(0Xo)_+Hs}Qt#C>R9gEqOorTFaTi$g%`n8~xUUYLLnild z*@{mQI>qUhS8+MYj9evmE*WcdoNyF(WrRbk|pXz&~O zdS-cT>!09jaLJAVaDFZVDMa+=8LqBoSF=0*oD&1mI%S|WqX9A0_%U3SCVlI}4H_6U1wods9~;#l_w^gCuUnVl#JTr}N6> z_iR8BZf~#_F~_rxSr>D9OSlI1Y^zJm|0{FKq=zIKAw(JZ&)Pe5N%7qG*+?~=2zRPg z9)K1fbRK&>zB6AfZ_@Tw+VIjH>-(`JbMarw#t^xvFjslo!9>%=w%uO<4a+=NuS4) z0R=U}&C4gX$cOi}7^7eTzt1||=vznKfZW)6pWYVG_992AD3kUjFbMSleZzG63xGyW z8g($je`6Kg&jF5Uxx0=4ZZtXB*nv?H=>)Q;j8gRZz2UvQROhr0NdwyS~X=$*-y1l%Zl$SA&6ma%B#4-Wd#Fj&zj2R z5PDMWQlB~PPW&NzhpHG6xDoj5eN2-DXq&UuDTK<1CDpC{xHVf0MA|S*OWO9(H9_hX zY|~itf1B6_h~(wvcfYEe`{71z1jkv%6%lSl;<)mA2kdGAAHd*~wNRrxz&`Rsy-Owr%;{b= zKVitpOvSoM=l4ITMSy%!J9mA!v#pUo{q{W8WHd7i21_<2_B{anvg+!VtD>6EX^Ig4 zxZy9|kByE>fS;oRq+HBY>3knBE6g%EG{8oPD(q>$zq?+Rve+e0;yh3T!uf9}=&dFY z0LAsQq_3BKZ0ZBRxykD923Q#ez~Ul{lycs|suAW7DO&VeZeSMS$`NvgOU<*eu(q=? zZbFV-F}1zL5~rIpsU#tvvV_E(*F7KALY@{aVmv6zApHRQSjV_9PWvZR%XH5qTr#v!wP#s--J0c1|= z3cty$xz7Tiy46(R>NW=wd+U~mW58+B2Zn2xhL|kf@qF9>c{4N=pD7M-t2=aeeX#}R z@1Y5h75D3tJk<2I`vWpZ*}}FhQh4qR?l14BmunBG&Zmce94Fr-DMhwjZ=s!&t!L$z z{X=7_o;n{QC>Jq+6EOjj@=SCh&b1;`R%h(?80}FoG)VpY$LnrkCOnLN>~8P&icibj zKA;iAyH3b$_dSf{w({L!))bMjxQKW~*LLW(%WUO8lbx}yety6+GXkkKZjPqUv6vn% z8ji7aMKtmEn}H4pbNC5xrt>hNcd&U)HwfDK#zV=t{i@^nE!pd)CGDp`npNaG`1Kj( z0kN=mrL^RyWg#H)xiZ}{25499V1|&sKK@ZB%1ZJ#{)d;`2xI(N_V%kot3gt4qa&YZ zj-%H^n^XJm@V~y(!TctsQ7Y-1iqfMXPU-VJRdU^QnyKNb*x4{M2SoGmCdj?2<-K?f zwSNm!Z_E;nm!zEN@pv}7)15zvb7#JAvO4%+emR0bS)}2GWXun&6!)_w=$bi{tUr2AP!R_=BQAO-{*lipaXD( z8j^Z^b_N-qSF0K4;T)r30dS9%d{cL6sCZ;vXMXNJBuLjgcxVtp&Hu&LSw}?~wQHXk zy1QXOkOoCS8iwv}0ZEk-8A7^CN?K{8yStSdLQp~^q(edwkgl`mJ>OaDJ8PZ47+o%B z=HYqvzW04!*Y8qjv#1YKW%@9ki}x?%>YHXIJyX-vr-P4%lX*OJTw)6V>PHS7q(Q=I zSd0?diAOSS`yB1`;*@xdwCd?Cpvv1j?)?Ks{4*8c7-9~5x+gXkPUu7Q{LSP^jGRP~ zs_K0tOM}~3wp-s&;H57Ukqj(@UbyCwD__J6&`xvtAMbxJhRGRgV$|PsuzbJIk|p!!W*L`P>w+YbN&iw-|&XBJK><1SZJ7YX&03t6iOtMJ0mr$ zk$BE#XK~PPo%QhN!A6et%5lipHB-<2h)Y=2Yy=R=+mx(Gp+bom#MCn*{cnyUGejar zK!fe>VGC{;NownCLSf_xzQ`965C_d|;84bO1T@l-o_*JV%29SkpxWMcsMnv(MF`-o ziKfi5N6;3B8O2X;XUZw+o#t4aYlEvG%WgUIyC?$JxR~g;(_ZaZ_|zOWLz>#p(Ca+8 z5wS#S7%qb|(a$2;p>>zp z?4!T1p#B5cOf1sK+3Lz)5r-y&`NO+b2G$%zKir~E@_Gkx;ygrBSnvc@eCCMqoda)ET{q&T|cRFC9s#0 zGk2b{vT-Y{`5);%Xin3ASN6~(9d&rE$y)R1=veA4g~ttL0{1QmhF$04mO0ZNkd;XC z*!*PnJV16luJ3&TK=|+hjI+OSO(uAt$-7m;&z<=M?}in zud@`A#&kx2 z;@ua)1VAxcz)7Ix-@*8;g>36p_WHw`q(C)w^+`C~9h>!`R{|JotiF9v^BmYo@^>FU zyhHj|GX(APX_79x%71clEE3@70?j%m0WB+VvrQ(0GC?{Ooaj}7t<(0(yeD#BHL9?E-co4TAgG2(ua*0;Vqp` z_T`oLS7V7MrguOG)^;;tr63V9iD~=x6q6|-E0veX(;_XI=^!DpW~Ww4Z4+Y%AwmS8oeY6W(#3n>(Pb(-nI@{()Lag7U$J263V$Zt)DO z`QrRPEHD6!GfeloJ#A@Z>C0EZvjzSI+@CjeKd{)(5vaEEbx&Su?)fz8VA2O`r#-MI zw0rw!KGhWM4vA&0bRP%@c2cnN3?s`N zLD)C_&jLa3;UjR+u}+LX*p#4V_}GwYJzNuKj$n@&StF#F$6c8yu!v@Iz2EYUQ# z+ZuaHzL7rBdF`+i8k~&2K~Wy3-SAN)KTI5~#T$W`}w8%^W$;*FP}&SBGRmH-D>LP1c|qRc9j7(#Te;K+e0Ho&J{`*5w-qWoU)swYjz(q8kEey=h%7N{50jF7fi9zXHS8UYxfs}=&^!< zv_^>8T^po!DUeCVqO!>&5Cjxf%vf)ZY?gajOM7jmrkf6jvit^^PI4_j%vU715 z;#Z=xPFZE6V4^w&{mp(GP-eiW&R}ss)yp~T@0D#YD=WL&|4z@{eJ3qSa^bA%GA6w!T$z`c_0lS?w?<JOU)*XCd zv&Db!jhy(oh@xBqXwe6-ALuB?Qh(az&-qk>DnP-nogKUqcrz3@o$LbKMaO_2^NRf= zlBD5Gt8giDmzPKRZ%cMP4EA*kP4I;;pLIj5iY$lIa~?HW=Do-T)LIr40j+$KZb8ux za7E8wi=0M!HL8G;Z1LVGWDa%|ZREfkup6>#LCU4$NHoMmu#eb`9U{z zt{0gNTb&9;nP~01X~rLBeob5|X|oSd4Q$Fbw9nq`ODV=vyehh(88$0zpB^YG{FuzW z(Rm~^^7ve1{%PRZ+%owRxYfAo2C;6`d;s#z>i_FJhan013AHys+o?aPw3wB zyVR8J;3Fm*LrbU!1v*j5g;-Fc(K1kpareg;Su=ufM3a^3_P{L-M$b#TV~{=m!4m+jKl zl5Sg>qAH1+qUO^ta^Jhnc;|8d{xl8-XvMZOO>|a3I0Sl;8l9}cHV%{ohnFi5dq|b6bc`?ROA2LQp>w#PLRb3WJMP9{T_qx{E z<`!k2(-^2I2}OaE$VMybfRu}8VnAz+Sa8p6ICtN#3qQPlIibaA4UAJr9Q zuI-IP5`8jjRe5s~-%Ptoy?*kTIzWWvJSghQ!q}LMZ~gV0<3C!h>oIhcnu`z?n(f2k zzV)Y@cdK;p``;QWU0z!CjD;14z0XB0k}?caH=3JR+cbDpd_G0=3#zA8I+==z6A0&* zEfy_g;;wBiWz8jn-DH6Q_umzZHn@rB;nJrzSkR-iP|rWgl|4{U zQ5)`YxiJ;F0$$4RS4w4qT*^Ra|LkD7hdW!$BMm@#cXR&+S)r(oR{^d8bG1WJ6ekn+kmAv+0?=3zFdX7)c|G-6>qKR$P>gyT5S z_Az+*1m5lif(Sp$C(_XMy1|ZEe^d&Z-laCT5-ACk3DzDRW;6~qDH%Roej-^Ek)q!h z*&WF_z@(L%LE5Tm6?)=efI3*_8tbD$k7^sAPU;tnVL7DK<@pB};0e1;&1ySPX2z(F zAhBgvFmDM)!QZ7yO{8n#+rCQ@B`WwT>H57%bI>j!EsIo;WwiIRe!Y87HA z%Vi_4yctnwG8K|Q^{@3#;5oQu)3t&bWKc<>FlB@-=|ZnEX;S{hoNs=A_bkctvYTtT zyp?TGS+;;HX}Z0*e_ACaZ?@n~?ClSFxkfaB_587Uz;Q81FZ4J94oHgqAl`$S_z-&S zpIG0x3{`0L33}{vsTcuWyJ!g7j2d!t8~0{+L*buMQoRwZL_{{ zl%Z$|5Qp-#N85|_7!uU&MapHry~yxmJy>|yEhBoG@zm);2h7(i&u{wsZOsQ&^wm-T zglO+bM~(U!+Kj;1ZFXIOrOR@B$fDV3zD_cYjHju+^rXiQCiW}zyqow7kw2v**8&Dz z0|CJ!>P}JoLz6<`cu6JsI3jp{ghwVkg(%X zL24%sy#~lM$&$fxV9YaKl4B4vkeb5XMu}%3WK~z1uC3NSyU+KPyD7!ku@3vZk#}Yj zRHY8Bf5v42`3I$*#x#I!Zry{2)S~ilr%<4O6t~GF^BsF z^q%yZQqv~gm*sfm{LsGC;(ZHfb6qbMUAYqR`{aoh0NTOHdL{9W-lEMo)2J=ROvX%X zf;EJ7WPI@hjg`fyn9)Kbyyq`c-srnpCd!K>UDaAy5~1d;lBGN!nbrZE*b#S%LQ{-$ zA2RMdy4Jy%*uy!|Zb5@iE82LH-W!R^{kbpk{z(+TVYHw6R?YXH^FlyEC4k zY~+8N3dzv)8N*9`r2M-elxerS4yBH#?U~mHniCL_km88q5l^$CpyC4PuW($Im|_Wr zy*25VIVety2_l5nU+-$&8XaeuOYCE9#7Aw| zI0#dB<2NLCmbtnp@ zuHH+s?y#v&Z4^&-DH5ri={JEjJEzvJ5v=!48`zK<_NT&_G?u>!sJVIMMojYH%{r0=6gN}VQ5%rQ(q-g2KD+%^zK zR}DX~zTUDRn6~ai`LY8CFn?QvJ`>E(gfS4Cm)@jni^K=b;czFXTRIKB z@7|^A`(t&|loZHF>#w{N|4AJcdc@x`A;hRmylYG;6A|tuB7^$YX!8-iLhL(=^(|C% zj7+aOXZDHcwh_IwudX7oXUbix^pQHnG}4v?`yJuZK!DaeGu&_G{IlWZdtu1upb0&q z+WiZ2E=OnE6)HN68VwZVWS+<``&rtl6xQAuc-wv9-GTQ3Cf$B*Sz;clf5`VUcWKRD z0UJ`PV33dI$EQnPausi=@f1X=sPX7tdSh;M@=tdPxI7i884i!!7S!Rd8GB}KpY&7S zVF*B;)N^8;(_7w*FHBkh45+=*;sJjOQZGY=I}#r%D_!gpM52{X6O9+$K_{H;n^lk| zK4i6flP@B!T2&U3$W%-44-K9o`b6bjG5zD(7=wbm)5Tt=kpj`b`>`KAvQ9&_1fh@D zQ7@>aQQ~HL_n9ubypBSyGFA9M4?WGN#P@W!YUR(^C%x$S(_r=9Sq7@}3**85VRPS) z>w-A&Ugx`3+TKl(*z=$J*GGS@;l@Kcmjs;|cE66E*lg?82rB%5C<2NaH&gO~DNm zEan-pk+KAj_|00iJC0XjpieRgtO;XeG@~b6VXm(5Cl&#VT5bnq5@B!jqYS6EQj_h; zkG1K-7n7!CqWitqixkZ?%JyrxOlVMtae*}LQx=t78_;jo$9DgEYef?EP$t0Q#W2Bh zE%B`qNr;3q-6si&unwuw0(!c@x`*%B**R~%pmOwiIhSdF&isRl9`UM7_+uYRF{-lL zn1W>Ymmd>>>Z?xDS$23C4PSl`Jsk()tjlM;>Ot39C;yV}<+)yw%C&twCYsxiQJ-jX z94O$!o6{b4z0-gDwuQ3` z+IoAxAb!ZfUG8-6@+yM^J6awmDUl=3nw1kpy18WjCL$@zwn$2z6k6S*}O)}0cznU)?>#ijtQP!01sBDdvX2G__SM4gql4gG83 zgmMb}2m;Ge0*9&K$T|}~Rj~08Z%irXe#*KOBjvwgm+2GFn>1C}y!(X%-&{L<`To6( zU&z?BkT8nUR}yUEyzIZP*)8Kxo>H>Zp}^9f4YIs5T?ABPwn50H32sCEajea1iky=T zFQr$SQH@2_#z(Ap^4GDloXH$Uli!b4X|cHgOQw>~!kYU|K9~oJ7~1&u35vopHEgrO zFD^--{zf|H#zyDOkv|s^DmwyrsB_!AAl5kWl|=j(Wm!<)(Z`Ti?65^rR7c1+hG~>O znOHou-`ek2DPS3MP~~sKu5%0Db|9C}On}#Ef1%|$7L)DYsILN~-A0SarcK91n%T5I zE=f%WoKX(cs)hur9J7iZ;-358cqsQ~Yi3a-5q8E3!XdAg69>=^U04dI%3(AdgT$7{ z3^wlo%BB@_+p58G7Xu@`=cJYzCM-+H*^F%}F(mIMFc}72R~eLx6OaVSKBp2^3)BFR zY|58k?1N5)aFt2WdhrO>yb?D_YczeU$~|vp|`Gysc|iGSp0ryatx+DR3~^ z*icy#c=<3&>xE>elM*yGh$K6Rn(V9AzaDPHlIpm8@6y}=!EW5Q@N+l%(Ri7U&;~JP zwbXTVBDlXVEPex1-l4KV8_)#3Tc7lpUZ5H4flBSLAy}ouX5SG1S;Gs}+-L`A|2T0aj-yfo5wIzYb6bB7OMmszwXk|M(%oB|i z*_5ie}8T&CJ8#niCS#zUgB zJ;9EWh&dC&O0C-l>-U+n%70t>eu%Qy`~OS;&axYpDDNfS2S0j!$Zsf%1GPXfE&CtJgI-Ozb9@0g*G4|EB~AOw-?vOnYRFnjK+75e^3v~8r?H;n8HJ`CU} z&L^tUBYbF%1$w7BITnz2vnBBiFe2S*PxZzi3^5i)oeQzi|>xXkxGoR%Sr)u*= zEYZL8z6ln>%g#N`phJ)b1%E($OB8Z5t*12%!ZHC|CV&(U?aM1>GlV1DbMA)_)F zA3U%I4v*9Lq)6xeK5=Ch)#;`a?&EDcX5vc96uR*cD@r`W)6j(E>=3&`lP;obtRj$y!Jl++sp0TQ8oT9R-i!9#lvE>o$S)FNQrAF&7&%HKY+!Dq#q`mL9(FiRL}Xl zoUr>zZB>t;UG-KwN}^~K@4yof8TyZO*jlGUn457cZvd_4kW5SBlP$F!CF#2_(IGeA zm6&!WCy8yN1o|lys4&v)a=_%06C8AFF@LbhCWMYZG1kF?)9>YONT{QZWeBMmP4`o@g4FF!ydi6`cnh)oF#5 z9|gPoq*TsU?4`@Go2)fpg((_DK;?pt6f;BA%Y{cn!F%`A7}v?B{)X0A*Ir<=&Y@jH zkLV+L(cugRlESGAU|*#Dz+}3LEdr~4qukYtE7OYu#|ZkZAcCfXd?_l!Ci5C(x|aX2 ze(v@AQ=}l2XziU(%k_eIg-y(O$@u}MB%pSl(?K_BQKIccxTa7iYWi* zplLi`4Ayg`mK_W<%4g7+p5)jFz`s%dWMUG6MS7L!d?XugwDNF^Q4pm~W6B^$-bomS z{Eb=Ig=DbicKNo#XmZc|tci11h;CdE*#;b{DJCSKB9o-!e6%aeHJq%pvIHC!n_#@I zlZkF&5yfenj3R0H{lS~~Wip)?Y89_VGQ)4BJEv`lnd5qRm0c+ItO!3aStPQ5AOD5} zw|_fwpW^`Dw1n1G`Z+mdG93DhxOJwjRZ4nE{K5BgIdU^UCZ)W3k~bk{@!~;j!3HPT z4xvzB$yuE{DLN^O4b z>cM0GjbyXk`do$lP=423ckW$^-h5C6&un(N4f3|IPcfD<=KXsh>veQESzF6<^KVMa z!74~Xu9z8NTLn!Z(sD8W#H%DUd92Nju=jojZ`ajKaJxDy`7=<04kPRhzQEjQbuf@J zVCApbov5hBc&q#b8{Uur^$LfvOR6SHr6$-BK+}UqSEbK zgj3JYxjoxohqkGEkT!Zmgp$sHq!Y1?l%RTw-ObSIpn3dM$>}G}gsH2u4cpBu==O%Q z)^$v^QwV)$2O4~oE19_XuxFVK?3FVVChr*u(At2Q2I)^o1M9IW5GAEhGXX);haur( z>eN!U>*6;W*m^OP{tx$N19^e*HnrN^R~h8a35_d8qPwPj+!lw_Oi16W8&;WqI#Z4Z zO||TK*1mpsJR>}sDCutKVfDw1`|S6GxEwjs4{wo z_sPTCo4~VRA_6?fpzXhMe<)W&;MDx%-&yYlbfgmX&gXX zp$L&f4FM4o;n>P_NC}kFudAx@;#>)DSs_DNV(C$L2y9SR`D&7qcibq+5Qs=cU0IB2#j}+KX0cFjRzl|l{IAGG_bS^3 zy#79F+bs|9As{w6!%3FWlLK&)2mGCXD?#!MJA5_V+-?IwSLl15BPKVN2PzQ%`q#fq z!SOiV@Ag!TK^_`}L2YPIKMW1MmvhP#a(oCxMIV3?R;@2$JKL%vtPv#dS*)i6V{Abv z*;;0FzaS=m$}@vu_SUzDOLD+r`w%P_I&0>nDufkn;5U|vfhUy)aGl)D>;}AZtoPX+U6PH2WrBc@&KmvY)p>2j$iAIGuFL)+Y)EuybmgxsEj}VOS z`AL`VJD6Eeqi@-kf(oHF(=XFvrb-@%kkPEzMnl=MBq0>~agCgVIPm#J+#19LSQO8S z6XLP;F=2#SO-2=l2J)Ka8LjR+q-_2jZ{=c%xD&(nKT0x7_p}3zxgOB@6vDR5J7fcP zck&;m*k2n6kHdsZHT_Pqz2JPo+=Bo(oP* z^#;!(KumgNxkG}iM~#et^qp5c03-OYC2eQV8Xw6Bh4Q*vYtKAuHnmweENTzAm(=m`Bf|CjfmjKgP$g+_<`= z)iN;tsUuEd;K8D(7)rg*Bk0B5bv8g@Dz)ttKiep;xXKOI@R9*=3@v0BXZn+u&A`lN zUPCt-a5hKao<={D>63(70bCJ2U-7lUlD%D4s!Mz(z`9Lu$Cc_ZH{t==J$h6x%kV%} z8Qz(Cu&)bv&D=<@kbP$Kn6*wZp2`=PW_zOQd%4G>QKT77w`-}#PIF1N&4hf{{Y-ZbZG%x2fJsyl{p+M;CW5~ z{;;8POb{LNb+Tf1Alx&uHs8Md1hmlDf$BD;-D4tcwXepACHffe`|WN;(eXh?vO#co z@7a#qyQFE58EeE#^f|f_1HRV(m||3W9aM;HKxLfxIkYBOGrL_mAV4UbT%gV!*em zF>ak%a$}ZB*r`>}VOm`t>K4r0R=;c?!8x+mXn%+RdCTt2p5O95Uww9Q^dLy?1y3ph zjJzt76jP?H`jm&}6?h5hg#8{Ic0Dyu)eEbFg;wi@(+JIC=XI^~zB&XA<&ZEKQO`xg!&@f)k4#G}9RMap+smxI3Do@=|Y!f}$pfgG{a`=Zq)CGbK zF7}jo9g6$pf61I^N`H;fcHZBqdv9mA7_@a?Zq>acw&l*u_4=^v>rxDgMROq2uRKdu z<7-n-CJd{hAN00XkYz^{Nt(R7mvpAErJc`Lh(Y$=Rg*FWaq#~&ySs^43#-T5D+f?wy;9haGAP+d7>{I86F96 zuUP_G_qg(gCP$klY^ShPjf6k`qvEuq{d+Y%EI?nTy?g*B!(2eoDi#ODI;`k(VCc07 z{^qe)wZ1X$7|Bt$dG388mDIA7k;k|S>TWZ|xOSc41`}%Q$I<5YhVmSa+u}ba9I$(B) zd_h^|t=D%+7cP&67i5Dvclca|4O!>DYtH57LMwZSP<(^VEgxL@yhjo6V~kHuhfcNz zJK}3+@@sNEz?1VlbJEh z$~}w#Q3*h6e~7r;Zb<-4E7e>ns=dd{{w?Ok>fXfm!r032X^9cMFd$@s~E! zjx{4`!?&@oyP4-Wh5)d<_%3^;6*Ls=znM$S2#t~K-#4!qHKMBMh)L=BfF#^9+gusN zY8s&o$M_~=2u${GG1VIXj3@uh+VyPxv+r(M2{MB!q{M5n)wR7*kbq_ODLbNV32pN5DjQm|EO|sr zQKpY-ML_&{$W&YVpRd9i`NIuw?-C&Uov`~+nGVvc?pjCeTp@E&G+f7VPnV$1^X@6Y z2gC`}#A309g-(-^ir7zP%F4{A8d(&DrOYh_^&0A&y6sHzZvs_VDhPTu9p=he{ZYkF zj7Fy8&90dR1yLwzm<^IQe+zag3~_H7Xh%WJUJ6CRYJMleSg_Wxmw;NrX|CR6$;Yvi zRC996Jn)vIZNHvdNGOXy_seE@&y&-S^_89Se+k!Re~{LGHZrG&35yeihS^UQ|HcX? zqF~GwYQ|Hu8|`M-q&G@hY&o=Z=z#`JykI3Qd#TWTS)aBEX{$uM`;GVqGGveuSUbnx zTb6t(gasJG%$7mL;r}_0Qwe@N;G(UnAVx$bXH^AlMGBR_WnFv@x6ZLx2Lpoq&;SGBk z$mF$R1RrJ_!^8?Z-K&``??SuhwI5~l-e}2SN{=NVw&+72MQaOoc74?MNioiJvQWi_ zAk1cqTKdWM)$!j4C}WM`$LD9+BJB^;$fBIZkhtv2(WIGK*N6+^5(Su^Xh{g^1d`HW=5R#(|G7 z?#8x%8v&tC+7^u5qZEh`j1#;0#I|hKm*$S*Zk9f_*79kBn z#FbcWY~lkT(`Ncm+~=Q0#&Ga}rm)Kr2UtuDV=&Sm9&{YScD{Woz-%C)R{DXvPYke< z!T)}{Cf@8Cy?^!EW?HuopTc9Ve&BfSWp4S}m?&>HoqPw7?!$MAg7Zh#XhAXAd9d7Z z&IZyhm4TC6$*Pv_LU#!Sl9%EEXX*!I9RAm+uizy6nyl@8 z&&3*EI!JNS)Izq8fJ>w}ol{jevG2h%&0NBQKk#1Sb8uFoh%5bIWPv<-Q^BOAXt4Du zVT5VPeBTBbmwa8)%5$1i&ol+>RPg~}H*@TF4>2^OaCw>!>#Z`DK@eZnO0ue=6>*R4R z0Wz69SmyLT<_oG);#DW|sN33pnFSa{J@ZFlzi-{WOK51q`s-N*?dP|~Z?)0?A!Hw0 z_Cf*q_#^4dC?*)3Nb+Bl`@!O7BfefGC6;KG#HF)A8o1qB%A?mPhQwc^VcY3!g3$d^6m=K{Js1vA8PB zgnIdzM~E3=p*^`DXwEX9Y(LQu($M?#D8KmA`qz4UBJ_F2pMXO=(gYGD=A|}ERp%MN zGx_JW%D52%Jd;m@8zFgc9DzE)YmYh0kgwi>6p14jQfswcA)VbDFwiOt)vvL5M%84g z#vbBy8f?pF+Nk|d(@&0K6@)=JEIo(gCU05;u|BQ7=V2GoZ&?P895Ze%0Y*#mEpXR; z0A8MvKP}6=V0~*Er*x}YJ@8NGUsy@$XAB59>ytpDvN z{hiffDZgDF<6EGrLYi&2v*d^kuXxBwpFI!Q&VUX=GQQ?Y8v*SwPhUjbH}cm^et&by zYab{smUf#~5-R#JM_!-Za)4yFnK82Ez@BH-t!t{MkJgPIT7EtFl>Nc&6O;~S(Y((s z`Jz8cryL5buG*k4zzjlXsG8LCYyoXS5rEqCmB-cz1fht)i$0rh2n|LXxFv8%Y&-5h zTeGbSx)5v@XKHGlyr6z$Y+1G!jwfwUbA}9W3Wr7AVo){eXt@W$D9kn9!w`El_vWjX z#AxFlyIJ7r!C}yz-+~qEfC;TB5>2a7Pxu(5fTmx)R+r?9hXT%*8khi7W~&#+mFF+o zup@M|kNtJufl^*S|Ic=Gv`jfpeKtQ-)qqX9C%PQL6Ve-E1nkc`tC8orB58wBkB1VS zKRGW513xg6{X%nrK8PJHnTne(O8#vV{WOGvaUCE#5V15DG_j~xyMV^z%{W z_ml!!d+$=RjmA!1QFBgqgsA`Z)Y+PzMgnU7tM9m@FM?;RbHOh{6|^TxPcVaA^L81(2cxjQcf#w8)Nw=C`eWI) z9^Y}=@n&f$$TEf=87IK|jODOfTqX6DVNSCy>z}D@-LEz3FK{V;Ve58KyecL*Q*sX2Or^5x zi>R^t_d8SQTY|nzk#|B>++*=K7k`-tP`{|vE|=tvPm#YhhDH!FRC@#^ll?^Czpc$v zP2kUckkx`ZOWqyUA`^<4t^+N%$QJnsW(Q)yQ!xp~Q%##fI=k4lic5UxUqcm>bYhu{ zT)GCnFiV%R1+kT?+$ye^{G=T|3V!+2xD!KGC5S#a05YG58L#k|cmUc8!dtW{C+LvJn!&~BX(Eb;#KCG{rq|~mPa>l zPM%Tn1q^y?PpU+o&%MMGY@(yYI~Lb_`-qX@!KRIzM&5weQtdT=DvPDQLsj0@z5~~lpEr&N8${{v7UVByV>fxKX z$k|bKm*>Y=`r%z4Dsofe{bE{}FdI$2>fn0{?(QT232``B4)QCg-`df71iK}`qzIl5 z04=>DlZGv^4ktR}+lzIJCO&PD=H%EYp?_`Z*;byk(90dqP~2YK^sQLKcLzj_ZZTDx zG#FZ0+I2=y^RUQrwR4qAH25%8upi7dJbNEHR*>%|<@Y66vN0p;t%V%T(*sRVGO4pp{~?Y2s!)=C7A6b&8yjme{K){&;^2$Z^3L6v>8K75GXNg~%B%7Q zBINmVj2Xy7`F?O(Is>PpUzvGFJ7aDoud^gVFa@}TOr|=`L93-4`sB;F;etHN)1NqS zmRFbEWymQ8m-d+(YqX3V@y4S|?knjzst|ng&*Z)`BNsTAb(I>#>nmLnEjNK zBbI0$R0@7gfnW0wI^uhLsX2>%Zh3S7cC;Hd%`U3*WBLAx*nF0@B#w^z;zN{NI)0uH0oul6`DgdrjIqgS#t?72^<6RxRQ2m@@_ zpeV=S{!{tY{?kAtAEA`}`$Q-Gx&u@q|3V?_C)3ub2`#@oGu?OKK6m|a^4#x{VzBYw z3m;Nb!Eq_IMQEMBTmT3Dk$!=6P;q`t%W1AponlnbnqHiGMeHy3$=#p;iuZ#(^5*b2 zGGi zzVFb*Mx-8%khIf#DU~wOH1r~pyWD6-_PK}1T+nT1RshuJ(zlDE(&I9!@HM%$b*S_AnAH$wm z6tiSQKj1X4KqU~!ASU&60R<{-=Bp2i!ROfb@nwkB2$u)t00cJt>+Q1n$AnJ$8Rb#r z+ip4!@2G;e-C4+OrUF9(XO}o5o}vFtuP}q=a_JEqh}9y>^~;a=udY?bHvmdX(oye= zD2z|3?^rH51e6nny}RLQWI(a93He6!7LpW(9J-;zxY&bqR$Bjz-|i4f{9|V8_KCv& zIi1yV<%w_%#Ow&`WP#ROyL#l0yy?4>-yn z2m2gNuX_zdyvtGW^+~YUdAi~rGIn>wVx^_kmrQV1GMRWJNeT+W|9*Ort73V=*@+0+ zJLVOL=;Kb!1#z7QU)PcmMFFEkT=*jt9CGVg7XFtOLM)M?f2N1gM7@(XC1c1pob)$G zzw6(A(?-F^L;nlh)}_UIlgvh8%5Si~lPF_?K2&mW`G+0s6ei@i9r~sg6)K}ZvQhgs zvr*#ma%RIe)#iBW9nN!*j5B5&9-f}iDyNdoY-u?o5}ttxjPcpRfiRx`EZ6!KX-0#zDFMcSX*SYSl>1_1@?Z-EqSY0V#2?YF=6b8py!$j@^q1Cyp?aLiRSO%t#P73c**Nj7A(Ett@~ zvb2}qoJvD`;h-q^QQE_TRzGR;@*x_*StRvW7EmCh=*3;kyPTG65(GPcG4 z@wj<>Ya5E6~)4EyRV8R8T$ zG%p@{%7r|Wgsn`1xa!x8BaHE^Mq^vk(CV=(!`2Y9+g2y3MxC@jCS1!0_h*6_ zm$si(?6n($-K&>P#Af$cPAi1nV-$d`T~M-xY(uc?v8Y8A%O1l*hvRw@_Gs|whHzEM81+R&>q6*8~mZ%!sSoOoPp3u%iu!8N+%ju|o6 zfuvhf5KN9ziE>=qUA`Z+9U$V~Kn(Z6xv0OW&TttN+5Ixyg}pXuXhOV`5cL*7r0?^m zjyNapxO=iAcHEzK@>xNC`-U3GYT2u+rgE{{-CorI7@l4DW3HtS-kD|Ukji`38Lq<( z;P2fXQ)+G*&9#rP*aQ8uu(?2N_)t8t?K#xwMW*`i#e_wG4H+)h$OHzc8CvH#Kx_l| zTLS>v8vjqkF8DhUL5V~U=r1qx_1Yb00}33n!BoR@2wo%gL%707=}iTa zf5Pv^WB4N9NTK1w=K?T0;Q?|Qz3L5^56zpW_MM=!kF=$Z0LQCf!W*FB1+BafBdxCM z8}*}n$)J};xugykI|NpqG(bEBaQS!O*rJN`bg-hGQim)8rW%+jSAZzBGhJY9m8B5F zr3a69=m*6Ftn^%o6z@xejovc<`agdS=idHfFSBTh5G?zpKVB73tuJn19?!%+Kqusm zAv)qeQsi$28!zg}_MUaPWzhRk7zr{KTK9i-^!$}2CE*Y_-0d3Ni4P*#pcOg>2jX(j zL{m4a{aWNX94kDFr}E-O&ixSC>SActi>ma*uK;w(56t+9AXbQ5HQ;DW1`Y-Rp5PEM ziiacc^u7P^j{1B%NIWf_T&%NE`h{iP1yFGWiu-d7C9%jRBA`H0>r-z|a>ql-@QIB% zwRNY36OEuH$Jk%Q*>trP$|?1eLy`>y9iY%5g@7(IAFqPN4`xe4V3kmPy(|Mc~B=xI0HMsLb;8Fp2koKus4! zSVE=mE(#JFyg>jCSgg4jgcRnuLZB6dH%~+fdsxx(>8HQkz?%7co`rhA={Geu$ECpN zYun=GS}H9}F|pjdBjPHNnfb?%cVODPw?e+V`w-E)F87eEEoXY!4)~@jII~6o@Q*Cg z(zM$bf`K

pH~nmIKz@h&1gY;h$wbu~pVU76o6Ad#VbQjVWeu9zz1`V=N{=D1(^0 zc2cZ)Xt+kc6NN%TAFJri{S=*DtpkA}{`sCln|NE>d$Nti&ck&0%nvD*Q^OKMNE_Il ziWfWSOH0T$em|Y#rTku}dX>$vg8xS+1flkHKE9bFMEARGhcb{3(x;B%-X^ldkMP=0LFCL2Ou z6yL4&8d;FE9K0C6X#tZBVe850y#3zs$WaQ$3C%d9VRe_4In+Dd0aO zPw~suN7fgquddrj6~IN!kGi}olLs%{8^H{3sQk-0#IN2EjDQr7jViWbW46B}1&jq; zbg+nvj1VXLCE!LT>yHFYwlzB7Rl%h4 zVC;}$SxMm)3ECxG=vag)8A6Vfq#IyX#9cjL4Z^&SHfF-hBfn0HaTML zMHGn1B@CxJLDK)duA`gdyIa1$4io?ro3^bC$7$mB3w&*fZf&FmE9=*V zjKR<;m0v7AY_WVNeURS=y7dcCA~GBOd&%&=>GWw%<> z0+)ChI~MI}!jt3jR||y`=glVR=lv7rVKU!Q;y-6}EqD1-&MM-j)Erf-ydZQlO2AWM zL!ArzY3+XjWSAflAR@yAZOzd(g{=gH!Iv}Ip`96{Ks;#Rqm$o(=QIInwEh9|y+%&> z1CH>_wZqto=YFSf#@zmeHk2ZMe=nG+!Dl}IO7aFmkE+!B-#_v|{T@Oumry4qaWeMA zp`rKn=imMM&+%4lP>H2tEfIl!;!R9G^%ta9O4<(BABWf_Md61nvL*Sc6<{D+bl$(Jb`@_VAHtWC$Nq zWFC*chMh&pA<--5=I>FY(DsZ8h}dXY@XKDXNlG~tM)b+T#%(ikYw>ApJ1vh$umyHc z+O~7{G$#M|mq3EhNCetDXtnB|`-HXTYgalMj;KSCF+yIMEXo`JSi_kY#(-C<32 z&$G@4AT4w$N>jShr6X1D ziszi)`M&$y%Rf9%2x}*6&ze0m?>p~khM>Cfjw=Y)`i)4jTQ%BG@okFzKs@%1b@>$R z#jJVIz72>_l9TMj*n!5dl8#O3ocHu|pMPu?((b=D>$PA@R}d8zOw2CpPL)s*(#@+-*D-v=RDTiq;yNfS>_RvEpOE`B^qw7t0q<)MHG^8!hjHvu>P$3Fg=Nhl4;_d)L6WT(|YHwC~gjR9BrgmHeh zDv)}Kf0iOVZEfVyaYs;-Ug9-d9;FM9wmgC~F0(T&HTw+Zg7MV;^mwmHu25rscnOW4 zshzEThQh_}O8ibl(=6bv+~Al>Uzr#9GW=^^D;QeB`~0KbhA?|9Ut;(glNzkVmJ)ya zw-CPI#Ke!jij7=VVPcKMg{p~tPzCguBe(=_mC2h81N^{@SY%KqGF&_=G6^`5{dC-L zCgKJdpf|)%Zw9Ph;}XPc!4wakcEzhny4<+#kc7l6;~24_Q%ux$gliA390%0JV9FgA z4FD^o?9xIyo$7WL+?vuIZ2RzD5^zsTAs5(iS6wEkKQ1M~%Tv#`P*yrskRhb|fFth! zHBC@Fv~`&kQ(^JI-KPAEUWLQZ2+4I-AgpK*17l<~w|`RG03s?Ao!!5mrdo%+3$^~m zSk6g9#9O4(_pq#XhGu7ZJYMVN^1^)A9jMqVrkPJ*4*eS)w0{0;bgGJpP;el~Eie5L zep%V)HRtm1MTzM1D}Zx(dvdwtr&hYz^v%mz$lNy0CffXM|5LwHje5FlvMrx&y0g)@^64YQ6|al@`ku7oxJqwtw~UtpgBHi!^Dcv26I2Y6}Usmc{fZbz7er~Dz#g2;&+ z6F5wx(g=eMj)=6w?{)2Nk~sHs9m!v|NIFFWo)I@Vmdy2yp-xNbX$+K})jEnlvxKv+ zNP;wv#HqtRW0@;nI0*Wo)<~3Su)2N%8YW6_2Cnmv;*4IUlYl!V-Jm*w_r~3p{)Rp% zrstGi@EGx30)_hR*WF{CDzOGC97L=v5{NIp_2#|vbz)48-~PtrN+mB;FGRfQ*t zQ$Th?Pqva7GhPoxIYySfmR-B)+t17s;)dW-iR#{%POKB)&1AAJvMe{P#btgz-}5qS zwbj+RGw5v%!sXV+m~pLhdZ(>3ch1V5TgaVj&V#*roHnhRwsNo`PS8WU)Z1_RKU{!w z{JM&jFZb2nrMi7YxRh;--MHpFWIN!LVx0JBIJiwTOj^fnV(H7PA3(a}#(RpE)P@X< zh{rS_a+kIQDGan8=pw=}1Rwch^Og5PBU`ype~zqZ+Vpm^l4?o?1w6RrK8qqLmgP7 z!_*E3`&n3(({q6@;?lT<>v!W0$E;-&ZU zuOy)&9`)#dt^toou2m(!k6Z5xlM55vOIlf&LiFwiuD%(5`5H&jY5s8a%^$R z8cOZ}Fc{LLb;`zq(aPO1gJmw<#p3#!w!F1asJGg?OVKd7n8k9`bFW&TZ|+gv)fMC( zMZ|t~6ayjDIf~&~O^5gHsNMZ^X##yTlsm}joS4z8^W4){c>@LZ9yzq1qNLMTYmqdy zejq8fF|)8%KK{9eL&5BXS`HMjk||O2+?Thd+#6}6LLu_uubn!PiR9d?3tTAWsn z>3?bg8Nv>KJHx_aO5?E2u_?q2Xiy!&UI*T4xc{6ZS1ffx)}(C&rKA}vbmWu&(7b`<=|sg_e$pd2)VmYZq;`mp`7d#^Lj|KB zVCh8Ap|kU>h;snPdcTYcTIBR&oHlFqq;D7&v|^`Z6X>x;uD0ILB;AKQ0pB0H+;3Y` z<;3Xwf;(z*gd5p(d4A{f?|#<=FxZ^hLt%rK3^@XW00LWLjf%#lKta{ipyVdh;kvG{ zpeQ7DC%b;Dl?>A&IN6Y8&8Ddt^$X!5S5m+A33k@0vs!%$y=HtaT5|_Q4~RrjHT|Rrtzx1;y>O^w7p)cnZGx{ zZt%l*bkG*rRc#P`_RsbV46RDBvkzEppQWA_RENU`l>jZ134eQxdo_Fa{@{>$^y>2- zFVpPKHLk9x249f*zZ8(iUKn2OO%3buJ|!1XGQjTlc9x>3FJCVVIq5a1^PI{0;akt# z?4W}e+Mr=A=0e^uXrFxkMB27{igr7DPO1&kJ}J_NOIZwN>ho3HlfO+q9hxn@44Fip zMjxO3F{t~GE!{ecH6OL#lXB8n1@dOEPoktSPR;q+{0s^L@I}D9Q+ssWdNC-MR(L%x zhd(kLW2kSc5u?_ALgat65Z7odV{|3;dUX_4E8c-)Xjw7vK!nL++bn7Gwh-A-PPKZ- zuF5)-(xSrRJrw_whE5o|Q+*|lHw7Zva`(^X>8uXS+1xAriRanaFH#RNDONX+yS&$N z`LOf#Q)2p>{)Ybao}1Wp)OBH6fUbvvJZbQqoM53xwH^%S93tI$V({HQ|0t0H=5eO) zq^ptfM`}Q;uVlw_`WUqctNiaO_8`5KW0OF#bbsm>J|j^yX$|jK;SG~Xe4~a~@bNR} z=$25y&4Z$_VLEn<&9Mqb_i5p&uTnnGmWv%meLP<;92o`1ak@OeQo##NOE~6OFuUM9 z@r7YyKm}m(B!F#M>OMG5(K6j7uTh$=-Uwoh2|DsZ*G@CjwvN0R#|v4&0QNgZ|xX;l1AkX1^IFh<0jFR6j65gWAeJ>r9^LXHR$&X#xxW_uKB!p`x` z&&m^&dCr(q!X1OqO9k2*m-4j~@+O*u1L7nP$d*g$V3%)PzIo(y@xyjf&FLYjqIpZ| zN{Z(>gz}3^)J7rd#&oP>;?_R_r`s_gyu-q2%aYg}Xu<)siP|C3zzP1T}QXM?%GfiXc z*NVHP??LlDsJzSyCoHnt&0#ktR_T7~6VgmnMn6nPa`qyL zKke+U(=Q7;$2BAnUIEkk;qXE%?NaW^(Z`cRB@+S=x}7BEV8aNi?X&`yaPjCKvZUR3 zWK(?A;UgATThc?SrAvSF&`oFMrk_DdKn_u1!Z$=3`L>S>WVdyY0g8d?!%WX=qRp~q3Z zZv=4NLGYUz6ar}uSSPa+^?l1*r_Ss?595*5-dbGh_eSQJnYdaT_Bd0DQDgHl zaZ9!^QQP5EIoenC)jG0dw?l&vm0IIcjDZRvvXt^dUwOv_+yY85P(MVBA_}0*7CG4%v+MHyigNbRO1Ah^6P#rkdYEcb{G;HkBYI54U z5sV;_Mp17JZ8I3G?8K3pmOx*~!O7$(dAM~Y>{6#Pwr2Dl&3pZeBK%BKS@c-xIG~W& z9nsGdQ-?XSL{i154Jlts6eQyBOsp1vs@B<6A)eC}E4K^Z?b{0N;gc4>%-{4nTvHxf z3jLhVYz?$^C2fn}k)tfOqX)F3xP&T-4mQd+-12HR(2a`I;d#t1ZjsN>Ot^SMh=dU0 zN_hHN2TFethYBcSGmZ-Jg|V`WARC^ZMhg-~Rk%htk{|XtnAsJp1XiJcSKE|uC}un%e%}ZL*PR%JrYRTfE3G|fxo8^&s%12ihyB4xrSuzdKJWJzxOe7A zrAogM_FEo&(6rtxGM1tp4Lc|k_6-1{cHt6*#5@}OqFYUfX05i8O(Y60f^EZvD)B(# z(NUEae|pyolP^)QF}6a>&_*#LLb`a#lF$O|0~Qo<;BV3q|lMK+~+^ z!{N)*P&s-Re@YV+6Ha@byXC2myo3sLL86&m3%_``!U_K(*5|zq1enlKBzWsSN49%p zF%*%K*vn0kW>9pNSOFdVYIzmg*5^Qj#`;V&-S|Y```JiH6Te}tjE|WSPVF;(Yf6DA zLqieo;iHE7o3K={ST21b^Q$Zo>@&r{cjt&tR&ic8Hb5R~yr9%kto($ws<`fZ3A3v= zB|tl|ZI%|$v=e^usq=}?k%J$V@k|^X^^Xx@J^<~XlYt8pJKAfzp=nX{jNFV=6SAzA zdH596rV}}F@p{yixR+6y$~inu{JMD2X-d0=vtTQ#b#5WUv*C;y6wVi(Ag!8^sjk1i z)h%TUr#`}F`m7_y^v_j{#lWp^L8F=w?Gm0%m_#)v%k%8pRdDOg&cP=r)E+hqd1I!h zx6w#g8Bgs})PqU|v2;2Hw`R#^Z7Ew`DIwqeo|^}Ao!3?^Kk6kpk zsnqrB?3$Bd_^(~1seJvWcj58vX#AtQCkMmngW_BPRMTxLzk%|u+37%J0(+Li;OKvD zI=JC;S97gsGRV2A&VUd8ZX{)mXpU{rBf#^@dHol1B{$F6iCCMjvhw3$*rJQp>In#0 zLz~=z$FDeTMu-(}_5ruL?94KjomUJ@7qc!}Cpe#NXhHFa?v(S8AM@2{;4(UX zoig~D$CJV&iN|N4p%X--SHIr}6Qy(58Va{mzc6I7 zT11XKr5%jEs9`kLx7vcz!hmL<$fn-*JLiA(&b$H&Lc!@Jf2g6s?0i zubuN~B=2MOVS_!|i;tagyDeUlr!I^MeT8)~#(m^qgQ63hac40r>?-`K zK)t=W^IH7157zF03Gx!EcCW$hzm*R+Xn@fT)r4)r!RT9*_QFUIlyUPhJ!P z9Wt`bac?QCVVeRtUOwF2WO6R&xEk2B+K0y93>~XrUY+-z`gP(5dUA_SxMe6!yYZzk z8TN8)YuG)Od$G#O?BrlS)+3WYEva}cDqZtY%gE~)_{tH`F9fDnH3;$^R@;A)4nC_LNlg2O3&t6dpGG^a_va@6r{wHB0&{b=gWBYZ?{<68ar z{<@1D+12;f-niciYqqR2fa>6Q2&PFqzj@w1d!=tiYCF67?o48NIY6wtUMHV*x(H03 zu11GC&vG64xnjK2+pE(_flj+(o$QkRXN<|cqTKxnR>jL6MP$@tRuNncZ<)kY)2QMY z{s1c9SG4cmx8Tgb_K}~1m0HL#+iNzq=I5iUF4xAW?v;Or2tZQ8q+!=0^{dbV^Hu^*$}sdE-)^(#6&uUVS8^7~sp zZU6cQYF4@d`)n~Eh7IxQkV;whBpr~xTae9vQlaEVnuO7YTV72`gey;^xPpqGJsB$_|byg=#$sXuWtZp|?C>jmggXkJzcSg?W zVLTLtjAwSPq6%$u;pO3DwJfSJkm#NdU+S(92~>zy2*IbE(+*V0<ZpinH6i^Yfly~G&?3~IiXnyXrJKZ#! z>Nb_Egt^vGVP>e{Op_F!)V+NVl;&*Z^?!a+yv#pzoxx}d&%70;eAzxjSD~_?zhDo& zu#Bt@SWtTHDq4C{uvVZ{y=!|)P9rL%nJjS)QQj}j#NPcjbEa!+&_*2cwFtaX$6apv zcUcVmu1nPR%_5rjUI!Kv%hh_zN)8#@Rn<0>K@~r2#hnmF4*==d^HqJ`zafDCP(J?y z!2JEYH63_jO8&g=r^fvs2#8eTHw*=Nzj4mW{6DdwMkXQ$;8xjnzut_!{69F(-(?CQ z026$H8SMj_8UOeLUIKKB_ku=Vwg>+}O&XarD8N;XLMyYI{{`D1twM_+w4yJ;IN;!?_JAQeYG8CueJ`)a93>XhO`ud)N@qmkRK<_1eXv=>Z zd}sVK(nb0XXWn=~xkBlb$t)^S1iC;kK-Ke`l}Om^?%L!zs5mQ_=xx)+vCK1wfheYaal>UhU*R1BkC-cOtJ+=>v)~z?(VC zzTsB%i^NX>Z(4d&sRuM6sjdn8fqs0Wb?#$q~#KqJIHR&H_N?)|6`26i|J{sOs9Erv`e`JU|BW0aPC}02?yp zc003CZ!@L)(a~1lkGP8s`gv53!&Idmhcu|Iu5sF6D9>8WwvlusCeOKti>N5d3GYlD z6&Jt0`DVZa6}u4~q=t1^l8_DQ@a(%HJfG>Acsuxb#{APXc zjn4$#gae=%uWlw*lyz(5vd0Jf4B+N;0Ot*c?`8QX+#Uun{x-)idh3VNpCZtHo7SUX zG!mGKqYCsf3uRTgy^LAe%@ct!7%KvM6g}bbr3XMLvY9F58yJ+OLegCVaFh?hEmv0& zr0^c!9}ENVN=KK!w20s`UZWE;1?J7BlzVh@`+UnS_^T!hYk%2`zXF@-y7*Cy#5+h@ z1C7?IK^t**H9)rv4coG*Mgl`q~;5{mnejO(1l8Wo%IU)MNK=;evIg_^V*`W8~|lYvsRzCPkv27&@mtwY>p;C z)Jz^%)4w89b?qda)htH2`$nT&!63x`7r{mLaD}cKI z$^VNl3nKzIWbFnD z$*F>l3K-gGSG!-O80kF%CG}686EL@~(#D)ay(w-7tO0$)_FYgJ?*gJ5@TW+X&q&|m z9#aZrCny>fjxp*Rb!@wJoqxOFvuEo(E4%a-jD7SwTzzYEw6}2}Z_SJgXree>e4c+I zvfT$5o+cfj>*Vh|SLn?9v;6h75!+LL1EC+7G1PComN*7+F~RiaI`@*qaHPz^bQJFs za3F3{v3m!kB31J0bQM?na-eJot>`<9hqD!tD`>Y|#UXAs;Ld&lO#|_JcoNq_w213b z4}|ur*_SPWOL9Q8$_<(n4=h>W)Jm&7BoVXA4;LcBC7j6ddPI>Hwt+4mpa@sPtbNo5 zMu+N~uoNPc>8B%VE6V)v&6*+#V~nUDQjXVS!apEP}!%%D`-ATy%OCSmeB;U5O@W@6OEKbxZ%FMeg= zc3~^J*$0RL&jL2DnN*PuW~dQelaLu${!^DK5zAPp+I#Gbd-9KFs9#o zg-^{5H`Q~}Muo}+F+Y*-w2$vT_+qk@+qz4J;sh`U0-Cld$KdfBM1EbJDWW@EfV+|X zm8+{2xP`JNFTet;+6G)c{GNx{OOb-4A%{IYs?=0Sv&-M2!LFA)w1@$;5{ABf1@}i| zr^*;u27q=$@p0a*w=VwQpZqKph(52zv%`;J+(GEQp&*V{Sm5I6ArGllx%lFLKs@Mz z0H0!b3!M|B_3u_(u&)1W1!^x#!Hw7PY%;l!gV~8B!BEfiVEa^kCl-g z^GV^D5*(dh>wC>H84dP1nweO~m~^cWt|f_kPWfsFp9Uw zi+aHW5f}6eP%%J9=xqH>Cnk|TNxNR-f&Io)jKyQxMbQJq`y6QYdEc2BVW|&X1t-1o zYeU^kHJGSTDKYztyN5GVQTqAMRBSg;EN(u7Ix;U_qnsvTqQrYmlTD#bWR6$7X<%Cb z%}S9<{>@Q{h+k(`@kWSYbuRdLIM@VDClOJL)j+ufP{~Eo-OAzerWnApUB#n{igWPJ#0!p6 zs$<5MPAqdqk-k=fm{Gw?8HG>6if30Nkh7y|-`gg&e#V1(&5}%L5L}yZ-`G?P2E@Kj z<%efHmn%_x`lJ=-VYUs$#V51`jNHCk!Z0j5W$*fpO=*MnmqUd7!Y=g)i4@~^(_!@*2Kt(6glCS?4J3l zQ==zSOt3}2MnIB_vEPvPOt{UtIDqUZerFvQQh)PZ-2e=2&H00ajyM6A{a6B9mHD)+%%FeJruy{DxiamAd8CMz!ja? ziw4qrAj~?yELRwRwnj`nZX(}|r$S6hj9%2Q>j-%qjqBbgZiak6B~LZH0f+>z=2IDW zaso7zpwI5T7tu1#Mzjg0$Z++HJ9mR z>CYH!gL9UOJBm^|!kXA!Uu@bp@XDE}7kf7i?txPG+^XORBcuo2!4;SAiJfmcNi2mk?sg7)GF#Z@wf)F1WSi0c z@gA|C;(5=Uh|@SW7%!av!a0vw{c+h$JtLk;%sGL%`_0I8E-a;tO`h97ug&_PFoJip=JfGX>u}h>XC!J3m#DHOV}Z z7;ciI)@TvJN>dwqu+Eu6cy9UhUP~J}qrp#=Mxn$%tUfVHrH!0=(hAw6!A6!FF(Bke zqD~|+p=9IxDB^|!(ISKTcWfvP;?lb3Y*bB1(*TbCI1Oa(p(~`6clVptM`jSz7Yjh+ zk@*^Vm-EK*Kx4l;O07S93R{V^#_RlXg{CPGi;&|lA0FvWn8Sn@hvMBLovA9G#^bcj zNP{IQm%l$;`*6S__#{LLvWx_JDAVH+7-jqm(d)3K=o_b#sB`CT@A+8H9g(m5(!@D2c(w%vXm`lkAy~CN}*p^^sRIm1tNR5 zovbHM;_Rv3Z~5X_b*wF>c_)e*tttP^)rliO&tKNqx)d}H3tqVwbn|KsGYAWAQ|BC> z1-UBnJzPAll~l!DQ05UKo@`z`$rJ5JVTV#Jh7B?2MBdc8$1Ue$5LiKj@kYHeCgsK<3HY)XQy4xf0yT>QfhXd@nWeN2Z<-B znlklnJr9XD_Bd@tQ3rCO9S%+y!Mpuz)N;YZJ&P{&sHYjy6lozUga=De3D9wBfe!+Z z`%nb#Q6L0&L}IaD3)S&n=IZyua0@kC@&$UiUL#%+kl4h}GS<_z%#yepXE!M)SbUysGz)tF>7(M5? zi=!MOVEDRu`fj&|_)e>G=vdrR+)mhm3I-K7#QB3gg9;&m{DId?4Ab|dd%_&h%+>x| zq@*BWmIHDM^-SF~AFj1TIo)M03EzVJapARht`R(dOYKA1)11S3sr0 zC5Y)d`%k6M_f^<7k+q8V&puRKVX2`dgnQT4`?hb~NfjCYLqG_H+Qvn{xoR3%l3BjV zRT{kaPOG`Gw?~ilHmfxyC*Ektm0gBhPyYF|a-BuJrV;+^$AX6~bq0xYGSGB|FU3=Z z^qL;;cMiYELqcM{{6oKK}eCU6+P;y#8*ZI;$BXLoPcMud8qTLLyB05P`?V= z6oIWH^BV9)3r>TF)!=fFuk&9{HPH;(T^c&f@ksI&JV`T=Oa(`oT=%#?Y=zr5bHgb4 z%c(rBR;8?Nf4@<19ep`)`%AgL9hdyQs4S!vNoEi!uoap!^HEpY;1X}`etEt>@mPrT z9P}myS@LVI?xujN{%_F@gCKRbs!tt>|GW6LLSpQCPu%!;?%%JY!7_F6jr$h=eaZj+ z^?&W#f7Z7lErST%YX1A}{!<*RBz4D*`rSQ#`R~^|Ny2()$b1xd8vcX*ZB5F;tF{^r zO#b~kfr;dWp}ma2{QpsoB= 1. + depth_multiplier: Float multiplier for the depth (number of channels) + for all convolution ops. The value must be greater than zero. Typical + usage will be to set this value in (0, 1) to reduce the number of + parameters or computation cost of the model. + conv_defs: A list of ConvDef namedtuples specifying the net architecture. + output_stride: An integer that specifies the requested ratio of input to + output spatial resolution. If not None, then we invoke atrous convolution + if necessary to prevent the network from reducing the spatial resolution + of the activation maps. Allowed values are 8 (accurate fully convolutional + mode), 16 (fast fully convolutional mode), 32 (classification mode). + scope: Optional variable_scope. + + Returns: + tensor_out: output tensor corresponding to the final_endpoint. + end_points: a set of activations for external use, for example summaries or + losses. + + Raises: + ValueError: if final_endpoint is not set to one of the predefined values, + or depth_multiplier <= 0, or the target output_stride is not + allowed. + """ + depth = lambda d: max(int(d * depth_multiplier), min_depth) + end_points = {} + + # Used to find thinned depths for each layer. + if depth_multiplier <= 0: + raise ValueError('depth_multiplier is not greater than zero.') + + if conv_defs is None: + conv_defs = _CONV_DEFS + + if output_stride is not None and output_stride not in [8, 16, 32]: + raise ValueError('Only allowed output_stride values are 8, 16, 32.') + + with tf.variable_scope(scope, 'MobilenetV1', [inputs]): + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], padding='SAME'): + # The current_stride variable keeps track of the output stride of the + # activations, i.e., the running product of convolution strides up to the + # current network layer. This allows us to invoke atrous convolution + # whenever applying the next convolution would result in the activations + # having output stride larger than the target output_stride. + current_stride = 1 + + # The atrous convolution rate parameter. + rate = 1 + + net = inputs + for i, conv_def in enumerate(conv_defs): + end_point_base = 'Conv2d_%d' % i + + if output_stride is not None and current_stride == output_stride: + # If we have reached the target output_stride, then we need to employ + # atrous convolution with stride=1 and multiply the atrous rate by the + # current unit's stride for use in subsequent layers. + layer_stride = 1 + layer_rate = rate + rate *= conv_def.stride + else: + layer_stride = conv_def.stride + layer_rate = 1 + current_stride *= conv_def.stride + + if isinstance(conv_def, Conv): + end_point = end_point_base + net = slim.conv2d(net, depth(conv_def.depth), conv_def.kernel, + stride=conv_def.stride, + normalizer_fn=slim.batch_norm, + scope=end_point) + end_points[end_point] = net + if end_point == final_endpoint: + return net, end_points + + elif isinstance(conv_def, DepthSepConv): + end_point = end_point_base + '_depthwise' + + # By passing filters=None + # separable_conv2d produces only a depthwise convolution layer + net = slim.separable_conv2d(net, None, conv_def.kernel, + depth_multiplier=1, + stride=layer_stride, + rate=layer_rate, + normalizer_fn=slim.batch_norm, + scope=end_point) + + end_points[end_point] = net + if end_point == final_endpoint: + return net, end_points + + end_point = end_point_base + '_pointwise' + + net = slim.conv2d(net, depth(conv_def.depth), [1, 1], + stride=1, + normalizer_fn=slim.batch_norm, + scope=end_point) + + end_points[end_point] = net + if end_point == final_endpoint: + return net, end_points + else: + raise ValueError('Unknown convolution type %s for layer %d' + % (conv_def.ltype, i)) + raise ValueError('Unknown final endpoint %s' % final_endpoint) + + +def mobilenet_v1(inputs, + num_classes=1000, + dropout_keep_prob=0.999, + is_training=True, + min_depth=8, + depth_multiplier=1.0, + conv_defs=None, + prediction_fn=tf.contrib.layers.softmax, + spatial_squeeze=True, + reuse=None, + scope='MobilenetV1'): + """Mobilenet v1 model for classification. + + Args: + inputs: a tensor of shape [batch_size, height, width, channels]. + num_classes: number of predicted classes. + dropout_keep_prob: the percentage of activation values that are retained. + is_training: whether is training or not. + min_depth: Minimum depth value (number of channels) for all convolution ops. + Enforced when depth_multiplier < 1, and not an active constraint when + depth_multiplier >= 1. + depth_multiplier: Float multiplier for the depth (number of channels) + for all convolution ops. The value must be greater than zero. Typical + usage will be to set this value in (0, 1) to reduce the number of + parameters or computation cost of the model. + conv_defs: A list of ConvDef namedtuples specifying the net architecture. + prediction_fn: a function to get predictions out of logits. + spatial_squeeze: if True, logits is of shape is [B, C], if false logits is + of shape [B, 1, 1, C], where B is batch_size and C is number of classes. + reuse: whether or not the network and its variables should be reused. To be + able to reuse 'scope' must be given. + scope: Optional variable_scope. + + Returns: + logits: the pre-softmax activations, a tensor of size + [batch_size, num_classes] + end_points: a dictionary from components of the network to the corresponding + activation. + + Raises: + ValueError: Input rank is invalid. + """ + input_shape = inputs.get_shape().as_list() + if len(input_shape) != 4: + raise ValueError('Invalid input tensor rank, expected 4, was: %d' % + len(input_shape)) + + with tf.variable_scope(scope, 'MobilenetV1', [inputs, num_classes], + reuse=reuse) as scope: + with slim.arg_scope([slim.batch_norm, slim.dropout], + is_training=is_training): + net, end_points = mobilenet_v1_base(inputs, scope=scope, + min_depth=min_depth, + depth_multiplier=depth_multiplier, + conv_defs=conv_defs) + with tf.variable_scope('Logits'): + kernel_size = _reduced_kernel_size_for_small_input(net, [7, 7]) + net = slim.avg_pool2d(net, kernel_size, padding='VALID', + scope='AvgPool_1a') + end_points['AvgPool_1a'] = net + # 1 x 1 x 1024 + net = slim.dropout(net, keep_prob=dropout_keep_prob, scope='Dropout_1b') + logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None, + normalizer_fn=None, scope='Conv2d_1c_1x1') + if spatial_squeeze: + logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze') + end_points['Logits'] = logits + if prediction_fn: + end_points['Predictions'] = prediction_fn(logits, scope='Predictions') + return logits, end_points + +mobilenet_v1.default_image_size = 224 + + +def _reduced_kernel_size_for_small_input(input_tensor, kernel_size): + """Define kernel size which is automatically reduced for small input. + + If the shape of the input images is unknown at graph construction time this + function assumes that the input images are large enough. + + Args: + input_tensor: input tensor of size [batch_size, height, width, channels]. + kernel_size: desired kernel size of length 2: [kernel_height, kernel_width] + + Returns: + a tensor with the kernel size. + """ + shape = input_tensor.get_shape().as_list() + if shape[1] is None or shape[2] is None: + kernel_size_out = kernel_size + else: + kernel_size_out = [min(shape[1], kernel_size[0]), + min(shape[2], kernel_size[1])] + return kernel_size_out + + +def mobilenet_v1_arg_scope(is_training=True, + weight_decay=0.00004, + stddev=0.09, + regularize_depthwise=False): + """Defines the default MobilenetV1 arg scope. + + Args: + is_training: Whether or not we're training the model. + weight_decay: The weight decay to use for regularizing the model. + stddev: The standard deviation of the trunctated normal weight initializer. + regularize_depthwise: Whether or not apply regularization on depthwise. + + Returns: + An `arg_scope` to use for the mobilenet v1 model. + """ + batch_norm_params = { + 'is_training': is_training, + 'center': True, + 'scale': True, + 'decay': 0.9997, + 'epsilon': 0.001, + } + + # Set weight_decay for weights in Conv and DepthSepConv layers. + weights_init = tf.truncated_normal_initializer(stddev=stddev) + regularizer = tf.contrib.layers.l2_regularizer(weight_decay) + if regularize_depthwise: + depthwise_regularizer = regularizer + else: + depthwise_regularizer = None + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + weights_initializer=weights_init, + activation_fn=tf.nn.relu6, normalizer_fn=slim.batch_norm): + with slim.arg_scope([slim.batch_norm], **batch_norm_params): + with slim.arg_scope([slim.conv2d], weights_regularizer=regularizer): + with slim.arg_scope([slim.separable_conv2d], + weights_regularizer=depthwise_regularizer) as sc: + return sc diff --git a/slim/nets/mobilenet_v1_test.py b/slim/nets/mobilenet_v1_test.py new file mode 100644 index 00000000000..44e66446baa --- /dev/null +++ b/slim/nets/mobilenet_v1_test.py @@ -0,0 +1,450 @@ +# Copyright 2017 The TensorFlow Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# ============================================================================= +"""Tests for MobileNet v1.""" + +from __future__ import absolute_import +from __future__ import division +from __future__ import print_function + +import numpy as np +import tensorflow as tf + +from nets import mobilenet_v1 + +slim = tf.contrib.slim + + +class MobilenetV1Test(tf.test.TestCase): + + def testBuildClassificationNetwork(self): + batch_size = 5 + height, width = 224, 224 + num_classes = 1000 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes) + self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits')) + self.assertListEqual(logits.get_shape().as_list(), + [batch_size, num_classes]) + self.assertTrue('Predictions' in end_points) + self.assertListEqual(end_points['Predictions'].get_shape().as_list(), + [batch_size, num_classes]) + + def testBuildBaseNetwork(self): + batch_size = 5 + height, width = 224, 224 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + net, end_points = mobilenet_v1.mobilenet_v1_base(inputs) + self.assertTrue(net.op.name.startswith('MobilenetV1/Conv2d_13')) + self.assertListEqual(net.get_shape().as_list(), + [batch_size, 7, 7, 1024]) + expected_endpoints = ['Conv2d_0', + 'Conv2d_1_depthwise', 'Conv2d_1_pointwise', + 'Conv2d_2_depthwise', 'Conv2d_2_pointwise', + 'Conv2d_3_depthwise', 'Conv2d_3_pointwise', + 'Conv2d_4_depthwise', 'Conv2d_4_pointwise', + 'Conv2d_5_depthwise', 'Conv2d_5_pointwise', + 'Conv2d_6_depthwise', 'Conv2d_6_pointwise', + 'Conv2d_7_depthwise', 'Conv2d_7_pointwise', + 'Conv2d_8_depthwise', 'Conv2d_8_pointwise', + 'Conv2d_9_depthwise', 'Conv2d_9_pointwise', + 'Conv2d_10_depthwise', 'Conv2d_10_pointwise', + 'Conv2d_11_depthwise', 'Conv2d_11_pointwise', + 'Conv2d_12_depthwise', 'Conv2d_12_pointwise', + 'Conv2d_13_depthwise', 'Conv2d_13_pointwise'] + self.assertItemsEqual(end_points.keys(), expected_endpoints) + + def testBuildOnlyUptoFinalEndpoint(self): + batch_size = 5 + height, width = 224, 224 + endpoints = ['Conv2d_0', + 'Conv2d_1_depthwise', 'Conv2d_1_pointwise', + 'Conv2d_2_depthwise', 'Conv2d_2_pointwise', + 'Conv2d_3_depthwise', 'Conv2d_3_pointwise', + 'Conv2d_4_depthwise', 'Conv2d_4_pointwise', + 'Conv2d_5_depthwise', 'Conv2d_5_pointwise', + 'Conv2d_6_depthwise', 'Conv2d_6_pointwise', + 'Conv2d_7_depthwise', 'Conv2d_7_pointwise', + 'Conv2d_8_depthwise', 'Conv2d_8_pointwise', + 'Conv2d_9_depthwise', 'Conv2d_9_pointwise', + 'Conv2d_10_depthwise', 'Conv2d_10_pointwise', + 'Conv2d_11_depthwise', 'Conv2d_11_pointwise', + 'Conv2d_12_depthwise', 'Conv2d_12_pointwise', + 'Conv2d_13_depthwise', 'Conv2d_13_pointwise'] + for index, endpoint in enumerate(endpoints): + with tf.Graph().as_default(): + inputs = tf.random_uniform((batch_size, height, width, 3)) + out_tensor, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, final_endpoint=endpoint) + self.assertTrue(out_tensor.op.name.startswith( + 'MobilenetV1/' + endpoint)) + self.assertItemsEqual(endpoints[:index+1], end_points) + + def testBuildCustomNetworkUsingConvDefs(self): + batch_size = 5 + height, width = 224, 224 + conv_defs = [ + mobilenet_v1.Conv(kernel=[3, 3], stride=2, depth=32), + mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=1, depth=64), + mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=2, depth=128), + mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=1, depth=512) + ] + + inputs = tf.random_uniform((batch_size, height, width, 3)) + net, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, final_endpoint='Conv2d_3_pointwise', conv_defs=conv_defs) + self.assertTrue(net.op.name.startswith('MobilenetV1/Conv2d_3')) + self.assertListEqual(net.get_shape().as_list(), + [batch_size, 56, 56, 512]) + expected_endpoints = ['Conv2d_0', + 'Conv2d_1_depthwise', 'Conv2d_1_pointwise', + 'Conv2d_2_depthwise', 'Conv2d_2_pointwise', + 'Conv2d_3_depthwise', 'Conv2d_3_pointwise'] + self.assertItemsEqual(end_points.keys(), expected_endpoints) + + def testBuildAndCheckAllEndPointsUptoConv2d_13(self): + batch_size = 5 + height, width = 224, 224 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + normalizer_fn=slim.batch_norm): + _, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, final_endpoint='Conv2d_13_pointwise') + endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32], + 'Conv2d_1_depthwise': [batch_size, 112, 112, 32], + 'Conv2d_1_pointwise': [batch_size, 112, 112, 64], + 'Conv2d_2_depthwise': [batch_size, 56, 56, 64], + 'Conv2d_2_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_3_depthwise': [batch_size, 56, 56, 128], + 'Conv2d_3_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_4_depthwise': [batch_size, 28, 28, 128], + 'Conv2d_4_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_5_depthwise': [batch_size, 28, 28, 256], + 'Conv2d_5_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_6_depthwise': [batch_size, 14, 14, 256], + 'Conv2d_6_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_7_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_7_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_8_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_8_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_9_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_9_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_10_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_10_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_11_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_11_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_12_depthwise': [batch_size, 7, 7, 512], + 'Conv2d_12_pointwise': [batch_size, 7, 7, 1024], + 'Conv2d_13_depthwise': [batch_size, 7, 7, 1024], + 'Conv2d_13_pointwise': [batch_size, 7, 7, 1024]} + self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys()) + for endpoint_name, expected_shape in endpoints_shapes.iteritems(): + self.assertTrue(endpoint_name in end_points) + self.assertListEqual(end_points[endpoint_name].get_shape().as_list(), + expected_shape) + + def testOutputStride16BuildAndCheckAllEndPointsUptoConv2d_13(self): + batch_size = 5 + height, width = 224, 224 + output_stride = 16 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + normalizer_fn=slim.batch_norm): + _, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, output_stride=output_stride, + final_endpoint='Conv2d_13_pointwise') + endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32], + 'Conv2d_1_depthwise': [batch_size, 112, 112, 32], + 'Conv2d_1_pointwise': [batch_size, 112, 112, 64], + 'Conv2d_2_depthwise': [batch_size, 56, 56, 64], + 'Conv2d_2_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_3_depthwise': [batch_size, 56, 56, 128], + 'Conv2d_3_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_4_depthwise': [batch_size, 28, 28, 128], + 'Conv2d_4_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_5_depthwise': [batch_size, 28, 28, 256], + 'Conv2d_5_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_6_depthwise': [batch_size, 14, 14, 256], + 'Conv2d_6_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_7_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_7_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_8_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_8_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_9_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_9_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_10_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_10_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_11_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_11_pointwise': [batch_size, 14, 14, 512], + 'Conv2d_12_depthwise': [batch_size, 14, 14, 512], + 'Conv2d_12_pointwise': [batch_size, 14, 14, 1024], + 'Conv2d_13_depthwise': [batch_size, 14, 14, 1024], + 'Conv2d_13_pointwise': [batch_size, 14, 14, 1024]} + self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys()) + for endpoint_name, expected_shape in endpoints_shapes.iteritems(): + self.assertTrue(endpoint_name in end_points) + self.assertListEqual(end_points[endpoint_name].get_shape().as_list(), + expected_shape) + + def testOutputStride8BuildAndCheckAllEndPointsUptoConv2d_13(self): + batch_size = 5 + height, width = 224, 224 + output_stride = 8 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + normalizer_fn=slim.batch_norm): + _, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, output_stride=output_stride, + final_endpoint='Conv2d_13_pointwise') + endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32], + 'Conv2d_1_depthwise': [batch_size, 112, 112, 32], + 'Conv2d_1_pointwise': [batch_size, 112, 112, 64], + 'Conv2d_2_depthwise': [batch_size, 56, 56, 64], + 'Conv2d_2_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_3_depthwise': [batch_size, 56, 56, 128], + 'Conv2d_3_pointwise': [batch_size, 56, 56, 128], + 'Conv2d_4_depthwise': [batch_size, 28, 28, 128], + 'Conv2d_4_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_5_depthwise': [batch_size, 28, 28, 256], + 'Conv2d_5_pointwise': [batch_size, 28, 28, 256], + 'Conv2d_6_depthwise': [batch_size, 28, 28, 256], + 'Conv2d_6_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_7_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_7_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_8_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_8_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_9_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_9_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_10_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_10_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_11_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_11_pointwise': [batch_size, 28, 28, 512], + 'Conv2d_12_depthwise': [batch_size, 28, 28, 512], + 'Conv2d_12_pointwise': [batch_size, 28, 28, 1024], + 'Conv2d_13_depthwise': [batch_size, 28, 28, 1024], + 'Conv2d_13_pointwise': [batch_size, 28, 28, 1024]} + self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys()) + for endpoint_name, expected_shape in endpoints_shapes.iteritems(): + self.assertTrue(endpoint_name in end_points) + self.assertListEqual(end_points[endpoint_name].get_shape().as_list(), + expected_shape) + + def testBuildAndCheckAllEndPointsApproximateFaceNet(self): + batch_size = 5 + height, width = 128, 128 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + normalizer_fn=slim.batch_norm): + _, end_points = mobilenet_v1.mobilenet_v1_base( + inputs, final_endpoint='Conv2d_13_pointwise', depth_multiplier=0.75) + # For the Conv2d_0 layer FaceNet has depth=16 + endpoints_shapes = {'Conv2d_0': [batch_size, 64, 64, 24], + 'Conv2d_1_depthwise': [batch_size, 64, 64, 24], + 'Conv2d_1_pointwise': [batch_size, 64, 64, 48], + 'Conv2d_2_depthwise': [batch_size, 32, 32, 48], + 'Conv2d_2_pointwise': [batch_size, 32, 32, 96], + 'Conv2d_3_depthwise': [batch_size, 32, 32, 96], + 'Conv2d_3_pointwise': [batch_size, 32, 32, 96], + 'Conv2d_4_depthwise': [batch_size, 16, 16, 96], + 'Conv2d_4_pointwise': [batch_size, 16, 16, 192], + 'Conv2d_5_depthwise': [batch_size, 16, 16, 192], + 'Conv2d_5_pointwise': [batch_size, 16, 16, 192], + 'Conv2d_6_depthwise': [batch_size, 8, 8, 192], + 'Conv2d_6_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_7_depthwise': [batch_size, 8, 8, 384], + 'Conv2d_7_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_8_depthwise': [batch_size, 8, 8, 384], + 'Conv2d_8_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_9_depthwise': [batch_size, 8, 8, 384], + 'Conv2d_9_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_10_depthwise': [batch_size, 8, 8, 384], + 'Conv2d_10_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_11_depthwise': [batch_size, 8, 8, 384], + 'Conv2d_11_pointwise': [batch_size, 8, 8, 384], + 'Conv2d_12_depthwise': [batch_size, 4, 4, 384], + 'Conv2d_12_pointwise': [batch_size, 4, 4, 768], + 'Conv2d_13_depthwise': [batch_size, 4, 4, 768], + 'Conv2d_13_pointwise': [batch_size, 4, 4, 768]} + self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys()) + for endpoint_name, expected_shape in endpoints_shapes.iteritems(): + self.assertTrue(endpoint_name in end_points) + self.assertListEqual(end_points[endpoint_name].get_shape().as_list(), + expected_shape) + + def testModelHasExpectedNumberOfParameters(self): + batch_size = 5 + height, width = 224, 224 + inputs = tf.random_uniform((batch_size, height, width, 3)) + with slim.arg_scope([slim.conv2d, slim.separable_conv2d], + normalizer_fn=slim.batch_norm): + mobilenet_v1.mobilenet_v1_base(inputs) + total_params, _ = slim.model_analyzer.analyze_vars( + slim.get_model_variables()) + self.assertAlmostEqual(3217920L, total_params) + + def testBuildEndPointsWithDepthMultiplierLessThanOne(self): + batch_size = 5 + height, width = 224, 224 + num_classes = 1000 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + _, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes) + + endpoint_keys = [key for key in end_points.keys() if key.startswith('Conv')] + + _, end_points_with_multiplier = mobilenet_v1.mobilenet_v1( + inputs, num_classes, scope='depth_multiplied_net', + depth_multiplier=0.5) + + for key in endpoint_keys: + original_depth = end_points[key].get_shape().as_list()[3] + new_depth = end_points_with_multiplier[key].get_shape().as_list()[3] + self.assertEqual(0.5 * original_depth, new_depth) + + def testBuildEndPointsWithDepthMultiplierGreaterThanOne(self): + batch_size = 5 + height, width = 224, 224 + num_classes = 1000 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + _, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes) + + endpoint_keys = [key for key in end_points.keys() + if key.startswith('Mixed') or key.startswith('Conv')] + + _, end_points_with_multiplier = mobilenet_v1.mobilenet_v1( + inputs, num_classes, scope='depth_multiplied_net', + depth_multiplier=2.0) + + for key in endpoint_keys: + original_depth = end_points[key].get_shape().as_list()[3] + new_depth = end_points_with_multiplier[key].get_shape().as_list()[3] + self.assertEqual(2.0 * original_depth, new_depth) + + def testRaiseValueErrorWithInvalidDepthMultiplier(self): + batch_size = 5 + height, width = 224, 224 + num_classes = 1000 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + with self.assertRaises(ValueError): + _ = mobilenet_v1.mobilenet_v1( + inputs, num_classes, depth_multiplier=-0.1) + with self.assertRaises(ValueError): + _ = mobilenet_v1.mobilenet_v1( + inputs, num_classes, depth_multiplier=0.0) + + def testHalfSizeImages(self): + batch_size = 5 + height, width = 112, 112 + num_classes = 1000 + + inputs = tf.random_uniform((batch_size, height, width, 3)) + logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes) + self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits')) + self.assertListEqual(logits.get_shape().as_list(), + [batch_size, num_classes]) + pre_pool = end_points['Conv2d_13_pointwise'] + self.assertListEqual(pre_pool.get_shape().as_list(), + [batch_size, 4, 4, 1024]) + + def testUnknownImageShape(self): + tf.reset_default_graph() + batch_size = 2 + height, width = 224, 224 + num_classes = 1000 + input_np = np.random.uniform(0, 1, (batch_size, height, width, 3)) + with self.test_session() as sess: + inputs = tf.placeholder(tf.float32, shape=(batch_size, None, None, 3)) + logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes) + self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits')) + self.assertListEqual(logits.get_shape().as_list(), + [batch_size, num_classes]) + pre_pool = end_points['Conv2d_13_pointwise'] + feed_dict = {inputs: input_np} + tf.global_variables_initializer().run() + pre_pool_out = sess.run(pre_pool, feed_dict=feed_dict) + self.assertListEqual(list(pre_pool_out.shape), [batch_size, 7, 7, 1024]) + + def testUnknowBatchSize(self): + batch_size = 1 + height, width = 224, 224 + num_classes = 1000 + + inputs = tf.placeholder(tf.float32, (None, height, width, 3)) + logits, _ = mobilenet_v1.mobilenet_v1(inputs, num_classes) + self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits')) + self.assertListEqual(logits.get_shape().as_list(), + [None, num_classes]) + images = tf.random_uniform((batch_size, height, width, 3)) + + with self.test_session() as sess: + sess.run(tf.global_variables_initializer()) + output = sess.run(logits, {inputs: images.eval()}) + self.assertEquals(output.shape, (batch_size, num_classes)) + + def testEvaluation(self): + batch_size = 2 + height, width = 224, 224 + num_classes = 1000 + + eval_inputs = tf.random_uniform((batch_size, height, width, 3)) + logits, _ = mobilenet_v1.mobilenet_v1(eval_inputs, num_classes, + is_training=False) + predictions = tf.argmax(logits, 1) + + with self.test_session() as sess: + sess.run(tf.global_variables_initializer()) + output = sess.run(predictions) + self.assertEquals(output.shape, (batch_size,)) + + def testTrainEvalWithReuse(self): + train_batch_size = 5 + eval_batch_size = 2 + height, width = 150, 150 + num_classes = 1000 + + train_inputs = tf.random_uniform((train_batch_size, height, width, 3)) + mobilenet_v1.mobilenet_v1(train_inputs, num_classes) + eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3)) + logits, _ = mobilenet_v1.mobilenet_v1(eval_inputs, num_classes, + reuse=True) + predictions = tf.argmax(logits, 1) + + with self.test_session() as sess: + sess.run(tf.global_variables_initializer()) + output = sess.run(predictions) + self.assertEquals(output.shape, (eval_batch_size,)) + + def testLogitsNotSqueezed(self): + num_classes = 25 + images = tf.random_uniform([1, 224, 224, 3]) + logits, _ = mobilenet_v1.mobilenet_v1(images, + num_classes=num_classes, + spatial_squeeze=False) + + with self.test_session() as sess: + tf.global_variables_initializer().run() + logits_out = sess.run(logits) + self.assertListEqual(list(logits_out.shape), [1, 1, 1, num_classes]) + + +if __name__ == '__main__': + tf.test.main() diff --git a/slim/nets/nets_factory.py b/slim/nets/nets_factory.py index bd8d7127a10..7c0416167d3 100644 --- a/slim/nets/nets_factory.py +++ b/slim/nets/nets_factory.py @@ -25,6 +25,7 @@ from nets import cifarnet from nets import inception from nets import lenet +from nets import mobilenet_v1 from nets import overfeat from nets import resnet_v1 from nets import resnet_v2 @@ -52,6 +53,7 @@ 'resnet_v2_101': resnet_v2.resnet_v2_101, 'resnet_v2_152': resnet_v2.resnet_v2_152, 'resnet_v2_200': resnet_v2.resnet_v2_200, + 'mobilenet_v1': mobilenet_v1.mobilenet_v1, } arg_scopes_map = {'alexnet_v2': alexnet.alexnet_v2_arg_scope, @@ -75,6 +77,7 @@ 'resnet_v2_101': resnet_v2.resnet_arg_scope, 'resnet_v2_152': resnet_v2.resnet_arg_scope, 'resnet_v2_200': resnet_v2.resnet_arg_scope, + 'mobilenet_v1': mobilenet_v1.mobilenet_v1_arg_scope, } @@ -97,10 +100,10 @@ def get_network_fn(name, num_classes, weight_decay=0.0, is_training=False): """ if name not in networks_map: raise ValueError('Name of network unknown %s' % name) + arg_scope = arg_scopes_map[name](weight_decay=weight_decay) func = networks_map[name] @functools.wraps(func) def network_fn(images): - arg_scope = arg_scopes_map[name](weight_decay=weight_decay) with slim.arg_scope(arg_scope): return func(images, num_classes, is_training=is_training) if hasattr(func, 'default_image_size'): diff --git a/slim/preprocessing/preprocessing_factory.py b/slim/preprocessing/preprocessing_factory.py index 35f8645ef92..3ab79a01291 100644 --- a/slim/preprocessing/preprocessing_factory.py +++ b/slim/preprocessing/preprocessing_factory.py @@ -53,12 +53,10 @@ def get_preprocessing(name, is_training=False): 'inception_v4': inception_preprocessing, 'inception_resnet_v2': inception_preprocessing, 'lenet': lenet_preprocessing, + 'mobilenet_v1': inception_preprocessing, 'resnet_v1_50': vgg_preprocessing, 'resnet_v1_101': vgg_preprocessing, 'resnet_v1_152': vgg_preprocessing, - 'resnet_v2_50': vgg_preprocessing, - 'resnet_v2_101': vgg_preprocessing, - 'resnet_v2_152': vgg_preprocessing, 'vgg': vgg_preprocessing, 'vgg_a': vgg_preprocessing, 'vgg_16': vgg_preprocessing,