From ec6aca70f70723818603d8057387458e53faf1bd Mon Sep 17 00:00:00 2001
From: Sharron LIU <sharron.xr.liu@gmail.com>
Date: Fri, 6 Dec 2019 17:23:41 +0800
Subject: [PATCH 1/4] Enabling OpenVINO: added configure `openvino`

Yaml config files under "cfg/examples/replication" were updated.
Config "openvino" added to specify the target device for an OpenVINO
model to execute.

So far, it is supported to execute GQCNN OpenVINO models on CPU, GPU,
or MYRIAD (Movidius NCS2).

This config is set to OFF by default, for GQCNN with Tensorflow
backend.

Signed-off-by: Sharron LIU <sharron.xr.liu@gmail.com>
---
 cfg/examples/replication/dex-net_2.0.yaml         | 2 ++
 cfg/examples/replication/dex-net_2.1.yaml         | 4 +++-
 cfg/examples/replication/dex-net_3.0.yaml         | 2 ++
 cfg/examples/replication/dex-net_4.0_pj.yaml      | 2 ++
 cfg/examples/replication/dex-net_4.0_suction.yaml | 2 ++
 5 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/cfg/examples/replication/dex-net_2.0.yaml b/cfg/examples/replication/dex-net_2.0.yaml
index f2ba3ac9..7852e06f 100644
--- a/cfg/examples/replication/dex-net_2.0.yaml
+++ b/cfg/examples/replication/dex-net_2.0.yaml
@@ -64,6 +64,8 @@ policy:
   metric:
     type: gqcnn
     gqcnn_model: models/GQCNN-2.0
+    # openvino: OFF|CPU|GPU|MYRIAD
+    openvino: OFF
     
     crop_height: 96
     crop_width: 96
diff --git a/cfg/examples/replication/dex-net_2.1.yaml b/cfg/examples/replication/dex-net_2.1.yaml
index 93391ea9..5fbc48a2 100644
--- a/cfg/examples/replication/dex-net_2.1.yaml
+++ b/cfg/examples/replication/dex-net_2.1.yaml
@@ -64,7 +64,9 @@ policy:
   metric:
     type: gqcnn
     gqcnn_model: models/GQCNN-2.1
-    
+    # openvino: OFF|CPU|GPU|MYRIAD
+    openvino: OFF
+
     crop_height: 96
     crop_width: 96
 
diff --git a/cfg/examples/replication/dex-net_3.0.yaml b/cfg/examples/replication/dex-net_3.0.yaml
index e8b9bca2..9e551f4e 100644
--- a/cfg/examples/replication/dex-net_3.0.yaml
+++ b/cfg/examples/replication/dex-net_3.0.yaml
@@ -58,6 +58,8 @@ policy:
   metric:
     type: gqcnn
     gqcnn_model: models/GQCNN-3.0
+    # openvino: OFF|CPU|GPU|MYRIAD
+    openvino: OFF
 
     crop_height: 96
     crop_width: 96
diff --git a/cfg/examples/replication/dex-net_4.0_pj.yaml b/cfg/examples/replication/dex-net_4.0_pj.yaml
index 0e970c56..f4f8a964 100644
--- a/cfg/examples/replication/dex-net_4.0_pj.yaml
+++ b/cfg/examples/replication/dex-net_4.0_pj.yaml
@@ -63,6 +63,8 @@ policy:
   metric:
     type: gqcnn
     gqcnn_model: models/GQCNN-4.0-PJ
+    # openvino: OFF|CPU|GPU|MYRIAD
+    openvino: OFF
     
     crop_height: 96
     crop_width: 96
diff --git a/cfg/examples/replication/dex-net_4.0_suction.yaml b/cfg/examples/replication/dex-net_4.0_suction.yaml
index 1d8d5d9c..3bebcfe0 100644
--- a/cfg/examples/replication/dex-net_4.0_suction.yaml
+++ b/cfg/examples/replication/dex-net_4.0_suction.yaml
@@ -57,6 +57,8 @@ policy:
   metric:
     type: gqcnn
     gqcnn_model: models/GQCNN-4.0-SUCTION
+    # openvino: OFF|CPU|GPU|MYRIAD
+    openvino: OFF
 
     crop_height: 96
     crop_width: 96

From c683ad28a752cbc5d20e4d3035f8aaebb1e414f6 Mon Sep 17 00:00:00 2001
From: Sharron LIU <sharron.xr.liu@gmail.com>
Date: Fri, 6 Dec 2019 17:24:24 +0800
Subject: [PATCH 2/4] Enabling OpenVINO: added tutorials
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Tutorials added introducing how to
- install the OpenVINO™ toolkit
- freeze a GQCNN model
- convert a GQCNN model
- evaluate the GQCNN OpenVINO model

Signed-off-by: Sharron LIU <sharron.xr.liu@gmail.com>
---
 docs/source/images/gqcnn_openvino.png | Bin 0 -> 28350 bytes
 docs/source/index.rst                 |   6 +++
 docs/source/openvino/openvino.rst     |  72 ++++++++++++++++++++++++++
 3 files changed, 78 insertions(+)
 create mode 100755 docs/source/images/gqcnn_openvino.png
 create mode 100644 docs/source/openvino/openvino.rst

diff --git a/docs/source/images/gqcnn_openvino.png b/docs/source/images/gqcnn_openvino.png
new file mode 100755
index 0000000000000000000000000000000000000000..b9ef140263fb80890a555e6931f76d105ee0d238
GIT binary patch
literal 28350
zcmd42WmKD8*Dgw-NYNGw6feb!I}|VUp|}+H;1n+|1yTx>;>8^b#Vxo)(6+d{gyQa!
z07>@U_kF&z&)C1t82is*3`Rn3k~QzOrd-!GXCgG!6^QVu@i8zkh?Ern(ZayM^uoY+
zV1tJVyn<OCfek!7@X%6_!>Agi-2?u3XeX;Ci-A!SM{sZP_u``uiiRE-7!)jje;+7m
zF`r^!xM(Z=Bdg<Uc9e~q_EdKfX}2=tHC0??j|A^GmD3PD?56F?n!;J}niKwP{Vj{i
zJtga5ME)kDGR1({PL`~I)C1PBOSC66btv0-Myg-+rfWD0=8+Rr=2B%mbFycgZFwL1
zSPm~j=u%$}kCFL&9Q24W^xN+@U>4xz7bM6lfjY;7c>Xt(jb&2j7bKncyH~vb%pXr%
zb4_Z{N9k|5W3d!Ro5dQ%a1ae>NSL497~=%-)4dT`+FP)l%wl)ceI)8>!4A=)OB=B?
zF2s?AiWA7iar6coI{Cs{I2yjcj%)Qfcr29mVVE&i@d&quShMggwFyp|_=P~py}D~O
zS?|yOB7AanzQWt@-%lx-KXlBSkYNiF;0v2>V`G#~Ki=H&G@vu=*?PJiDX=pruBv4|
z4&tT3>-JsW+xb*}suL<vhM(7c@tCEuHARzEKh?C!GLGDmi&6D^G)FdBIdiKvl^Fk0
zQ_i(s^(QXyWSWhf*HfwCk{#+6K6z4&%Cqo@FHar>S#>IVkn^;t_}6&J2S+r-%G)$Q
zxDg=e-C>p*CVy`$<{|vo8Xa7O44s}r$_Ft+{UpDRbklJ0$E41?kW}1%B2DPtajMdG
zCr;txTZDdR*dB<+*tAi6xzXzKfx;uu&n0%&f6=%bv*(kpnfkDReg-t7#kL)5_CeV+
zo<-$R4_-0rbW{N+!<5K}=NvOc7yqQO^qK_Qru@FrYI4#j+-6=ceuE$GKluHLWWKgi
zlbP`=h?W)~uoYOT)>ZkMJeI3I=lPlx2bYa}J!{#k_Q}EG!c^twggyU=&ha<%z(z3h
zY`1I8=`g)@-#+HlFMNy1xxT{@$B)GC6QAU3`l`k@`xC3@?}~UauX1@?ZXJcx!;=WF
zlDa*J;%={AnsE+QZ5lNw2Frpw2n~w9+EM19#VIxr(ItO`)Qbzna)0{1$g7IQdq3qz
zuox|jIcOXdw|e$8+57aX^MQKm1(oBrgCP#q@M8)NgS4Km%RldjG1DF<`(vBaefEOu
zRVj~%j{5Zu^srA+5?YGd4){f~G<~qB9F|*I>}Jda{2Jc_JHyaulwjC@wRD=`pFK~-
z8chu<ejUXeF3TqWEIsbOjzK_;RDJx%k9@zdTvDmNVy9*5=|t_|(4(ACJ*A<tO5Z=!
zb>rDK2t6j-6p9y7gPpI_cDD&tU+nk}n>ZTr@Xj)t>)Sl6ao2U|tTADX-ZJ1#kTd1e
zf3@xPmB6@gyDz}d-cUb{b38+-@^p+c*+2MYHZi*ukpbx4{5Sjfx0{Gq9_G0+d3?fK
zM_DWB(o-Qko2o8%O-|Kls#Z-q*-_n6C3uf5@3S0_j+X}4<|{42b*hFaySKTT-97Bq
zrE;>vT)ZfUx5hjFq@OKjSewr#8`yu*55%*;AsK*I3Mh8#SW#bL@F8}4BDeoc@at^S
zLV*w5F74Lj7ojbsr{N)u9ua>+@mroMFzg4u8Pk_qB6g;pD%CLhkZG@@(bhE6B~f1c
z^4GvE3=7zh6Kp`9NYYVg{iKfQ%&!h6prLV(tq!Jo{a~QMAs#aXW%S)(tmL!^PZMKI
zTyiwMb{4-)j|<Nxl6q=x#I9DDc}SRhoIMo2P8eXzRCFpSs^CJr<iB`e{wImt|L>hd
zE=+CllW+**ZL8TuRHO^-sI4AFB|$B>-ArsF@dR1v{JJBrTJ-+87aj^17j7n0jS?W=
z&rgOQbI=xD$cGqWXj_cdu@f5THh=OOH1Fluz}(*Z`Obh%=x<0G3b2VpL*Jfy8Yn(+
z?7EHY`Qo(UYeOjDn&3F|0viX}@Vds#UuAxT#uk2YeqYm*s~ofZZTM})D;wgm;o2{i
zAz`D!K2L;oT2yZI&Gq<yq)|o5+o}qI0?d0@CZh9hA6Il6Cm1#ih6?cg-owwUUj0Y2
zG_Tuol*={8#j&UNw|z~D#;YyK7o2m;4V#|q2PKs$BPPXC;VqaWT<xrU3O2DYfp<zz
zU-*i16?_sl50qDur>d45EKsV_je7d*at@PaSnxuNf9X*}0bfn)?AWFv(SXxKXQK!D
z;lKGN)Nl!&c#jAZ<$t-B&&JE+RU(X|f)5L~PCXm`>eLAz-`kGf^2g&~`kG8&p;P_A
z`p<Ypd}wQ>gT-iZ!=O(;b%@wUTkRLLooV<EhAPe%<En{W_B~aO+#Z&}>Xdl}Cd?&{
z9oxNJO5`y%D%&P`t?4XNj-=c8^|cjdmvvmj+CQBbbBlZB@C~qLXSO`kC|j~H7aIR{
zicmjWe7y%Mwwi)dy^`F*dPA+HCgGb;>Ymg`6gRcUQgpL#J#r1>E9Eq(!D-WrjIFcH
zPqo6etq|I1xM*b7(lHxtvb|Dy=<(BC>S6MezCo|IQXYKtah$^et(Gy}hy70Y0%`<n
zm+8HNBa^C4#0zCMt7793k9o*jB!#xkv)<a6nr=T5)~NU65Vo)#PPZv~y*>+MXY6v0
z9-(99v?&Wl><SStuXOLg!`8G075rznj=`kc>FfsBT<P6Isw5Zn(+1oJMecM73d3)$
zC1*MQA?<zmoOAX!-bm`E_=M`2`ys}Yg$kAkyc^s6l=)AF*b^pY7#=^{X+QrQ^jf_p
z6co$j13mC91+2}Is`Hs^NcEGu3QcPoXH*j6pWAvbTevM5F-pU?&HRRArpxHq@uLV^
z-=R90Z9=x;?1YkKT8;dF+_qfJ)U-+{uzw#b)s&ZsapKc|B2M*UtKFQuntAL+w<;9#
zghVBu%H=sNZ_d8Is803c0latpMz~DxPUKo4Cj6B~?IS{j!+EPIg_$j+0;!JM(SBkN
zW0F5AHSxT$&~p=(CvYU(K1^3@F?*+HS|j8^?2#-MW7|4~=h4CChBJccK3GAiy6CCa
zqgv~j;%D)C_vP^XFvYo_xZ;z))}%4gxL2816j!kbnq}_krD5P1e6=Y88|xLniJSSc
z5JpxX(W!~e)bu%Z;VUtj9{jbbAckRYBmP0W0h8yka*c`GZK1bGi%hA<RuI$R?$Zm^
zbgWZfywAzyT0}jn>Mw^mzLoz-GtmDoCRfKkXXqs*e&89J>ZfVjD*Ud?K@Ui|?H63d
z!xc9EElO!h-XlFp8IaRzXR)fMWWyBS@3V-t9)5>XyqrBa&}4kQK|S&CAl=D`eLPPA
zvv4LB(V3LIWulap7X0=_#@pgIww{fj;}<@Yuk_GXAlAE_9thmw$L)Qe6*q{L*I;k$
zs#<vW9fdvo{22N8(_aIJJscWn@jdt+uQv?h1utM*n~B(_98|Q@XA6Hi6(AW@ZDPeY
z7%7|1GDxKuh3qwtiiGhc(sB$LaT@D?=XlEexj5wMgIx%AOhYUoIrhW3udD_)Tf;Q|
zdapkD8DtxEy8I;gr$m#twQ_Dz?~t!<c-N`pUerdui4WJ&Cf(iZv!*$Rr@RN|mTp*b
zep97BsV(7E5MI}yZYfUQ`p0J#qIkVo(&j{6t-<!!$G3xHWju80WH!Rj=uHV~S@C_(
z^bh7o?ufQI6{GkP5BY}@IG+4(i?*~*>38lvP;(BZS^VBJwj@mmi_9}2s_jXqN_##x
zA1RwbWVwV-D^5@=IVNt+%^7O=hxPt>Lq%tgRcxOmt)XaI_r!36nNk|tRswPNUz05U
znlv6~I+x~>E&E%Xr;bh^EKvDWO)1|?)8eqNr@}_<X?0aV@;=6fP?UXG3rCNtdm7il
zXd=7rI;JhtWA6*^|M;9gRkdE?<KbRH{`sH&(^L`_eAfNX@ic6aqc8D|HO%A5UttUy
zv+&DFE)~*AzNj|B&N|}^s3Q0LXI7EVTyRd@VCT8K#fwrhAxqtGRbIE4IRO%Ck44_S
z#}y{Y4(6i?7eZk@4ir7_)-LxlWAL_SD|LRi7}bu(2eq)8lMWB?gF@4H`tSC}b)P%t
ze-az`rtD>I)LF_WYSgRh4o5V7DJZPveOx*v8;Nl_H-wR6U#0iN%r`~+$Cs(HUIJw6
z7y5$~ci%kiA@@&)T~yX1g?TCDJr!M>(JfWo(;r_`W(nFFg?S`9t<d7wW;GCZpS#w)
zq-|48_4?FJEr!Pc>s^bU`x7>V@k1+_*Zi8NCw}nvyZH>!3y#`o+_r?M*7HF^kFP`r
zp&7baJBtq5h2pOjZ299alF~!dH%~9(dffsMg|$R4-2c0meqB=oS1r`acvPf&T;F_^
zD584uat3@OoXNd)OAO3YWIyPX;^_5-Ry<BAhhhq1Mw;B=hfEcWjNwRt>|UKLrsJ+h
z_oOjom$StDGX@I2Nu%E<%;;#N4@>bF?Jx;_Kjmp7$T^0>&oa$qn+5`O)C*N9_)^PH
zYyiFYTpr{0zy@n}cT4H(PD}Yz+Vf{&Bwtgil|2HLnfL>~_nDhG)2H>DNR1)RVfYq$
zr|wb?(c@fLE!X5YJCVO6UpR#ikPnZaME9idf3^NL`#{-`tiH%2n|lX8fMt%-**DKN
zBP(J-0*7{g-81{cFsV_aaz;;Ah05r#SB|42wQd>B?17wTfq>j&+k}o`=mnqdmym#*
z|6~j^+sA2-pPvc|#t><!6qpxoAzLUr=|V*Wy<g|`;J&U7d60QzPQzaHrFbHYaqcl;
z#Y~<s&QOq?mt8+nGcc@MQfNmqjVx?IyA;QE*C+dJ7TyZB+0FN8+9<D2TJwE}iw!os
z<_T`K!$sDF&%EJ70~IgoA|F`ccrvw^Vf}8XqHV#nLAc}%y&kn4QTn0-SgP{Egr0-x
zp#`g$Y6_0W`;HV{D#qUz|Jg)G{+@PWYH7^`%ddM6)*+$a?q7qs&DL1z+B5X-LQOuQ
zlohS>prmq`0JV$Y0%)Eh#w{mfd>Bd@@6l^83v=yX$_M13(0?d^*FztZ`T@@$%PDUD
zeG?vId^qrrjG_O~|6gBBf>tfoX6C<)%rCiDiB3i-zk4eGE<Wj*BQZGH8CWTF^TGf1
z8vkESk9-bEtyFn8k=Q!gBL{HZr?QV9ly?e*=I1j4HWa|+z=ip2|37<Xrlh83+V1Pt
z`-KP;aMlmHU#H}kMV*GrwBM|LMBZL4wfR6e^j|5dsFa&^+#W!nsGl^hI;C1$_lOG)
zCa#*A8c4_WVy0^+0)|G-?Ocqc3+~M5g97i4nkGi~XUYqQNo4MyN4&i5#$}+D2{M`N
z<+JSjs&D2ezi0>z>QMID=nBRH1zyd_pb=wt$88#;zllVbKQRist_Ji+lJD*9l`Uic
zjG>t;P)c0w_&Am!TrM9)ajDPYpARKXkT5hfwE7wCcYpQ^d^?q^{MhGw35L4vC6QUK
zu?jEK#zS0AX>1t;p)PVEMh(h^Uu<EWfpZ{mhmPM~Vd}T@tyIt;PnvXK>r)yr_X1^n
ztvXvZC6YHAKVx+F($W^{?8>LJ%*!07OWZbMM9bWXt6H`*>~j2UCkvDo78c54Vd#xG
znd=i6^yAI(D?<&btEb51$Q;05Hha<q?SNI$C283-qOgDq;4XDXdT_E->E5am(yAAJ
zDbk0Sm3gX0(>Clu1Z{`ES;B(uDwXli>vEA!=bHlw<oI0ty7r>rFc2|~$YJY2l{QPy
zd1CJUCa|Kk*M8EpnxbNwm<Q|Z0vVh7;Z!lnzbl7YU2Kn#NW;J5&&XB*M+59BNqH`Z
zZkbM{(bd%)2)Yp~@wiXd``c`#artdnBpeQxLGEd|<AD)FpwVoJVgb3+^C4c9KgKRC
zTe}lK6wlTnYdIf}#oX3IH(N=6f-I`8_Gh&HWcd(HE8a^{yq#aC9-Lb4u6{h}U)Qz4
zWq+67*XX>YLxzLB4O!U-H(4Ry!AhgF{nr(H?%9niEsgXL9*;`z8grkCnz!f<KQ)r5
zJyA7s_@d+H45}63k$GPj#jJmlkA1lQt~f1W`ScA3pXt6iAJ$6-2l^K*K=?3r4_-|u
zFz7xJZ$2$Es99fZaMae;){S^Bo$ESX@{UCE^b-eVf%8%$N1NumVs$z%41RtX7!CNE
z)vs7Oe{94F7;Q`{b<6qbiE<UE#l5hIh@Q<ucGK3&(?7je=Q@f?N`@`&TM7(;)~jQ`
z%JYGDm6B4vd<IT_PrQ~o|HVATGE~w2`p>-PmIaL#{ZDrGlK13a%zv);Ra-O*3~oGG
zQj{gIz3UdiBs@M1QJ+rW>fkKG)*D^=LDsMfX^8}Kry}YK0xq4WB?}?fg12QvL`3kh
z(Tzh|_sg1_UnT4N8q``pddm~8`Qt57^8VYM(d@!!&t!tynYZ^LzA-Oo3_SOZ)a4Kd
zKuo9(RcPkPX=x=4dSGq2w(ha&io?ab0(dE&eu65$3`ZNL;4Fn0z93jNFAC>?s7==z
z)HCThv|vy6Cy$)ZlaD9;$G&nTIWP8jo4TmS9>n4wFiRiZ6nW518?4fxM$RU*P`g;Y
zOkGuVAc3W(UY9EDA=e0l|1KjHngI>}rMWTo`?oI;fQ9qfKC7jvaet8HqsTEZN*jE6
zs{y`ET-Nno--p<Z<#eD?ce)O$stpb^0)m1poP>a*PEkZX=yKD2`|bT5r_-a3=G#QY
zig(7PJSQg8a*N8v;5IY1NF}T8GVOXzGrfV<Kqss-`{Fjms#Wr-gr3<bjBFJ!5fk7O
zJO#zYj%W;~x6ztZvRY$yCnhQUcSA}2u!8LQ+(NC8@9Rsh(8{v@KF^a?Q=3qXA&A$(
zY<If_z1gzd>w(bFhdL&m0f@7WUsIElI$v<BKfPILaGa|$YS<r67aUbzs<;2~mVr#r
zYhNchgKT<+b0SXqR<Y3E0IHtGm#0;z?7+slov0xlFF;X+T5R)iO@2ytF>CB<s1Rrg
znmvDH4<k|UE*<!dsLY!9p-}F?&#xhkU^b&IiQc^{PXndBk_HdzJxjYDG8y{8jK4cm
z?5(LU^JP=bWJOvTbG~wQoW<+!Ez6nmdmZq&Ri%8veB3%N+Qd(~_XPK)v*qb%NV3?3
zZWi{3>a>?)AL>J%h~tJhuka3jZx@q)Pqt^VE%pLPm+L+G@==$8zkvj}(q7seX#e)t
z99j7?J$m6JK=?QkR&!}tO(EjCdY=iqf4iT;uFcBAf*lb>$s2?K5+%WBOvQ8=Fba&m
z16BO_n+btr#Ablqc|OP)f=dz`7e^wpL45B!Uu^;Dyxj+a>>~ni%(-dhdfqCo`)QLa
ztP0<B4JaFYf<kxeQM}%};}6)+t2Rx2R<M2gERSO)j=M_U#!%a|a<yOV$e>%I(Nh{S
zx}{?LP#LRLk+B&Nn|hY-?WH}EorCt(utsG1D<pXlo?}uH2z@)ZoocB)Y7^&@VY1U1
zs*L9C>>25V9ORA-o@KCg(YUrVdUaPA*6BE0B6UsrYHRq;wWMEtuqpZQ#L2O3L|2QF
z#*$A@KAUs0h^3=zY*@FM65RjiJB7%2cDrX+l>t*U18JO4pn7E93-@>)vd?Y~K7G#$
zC=SW6`~G2(C&3}+j(D$@%1<NvLTfna$KPttBjhlxh`o`PW<OO_ssja|EJlx=L_h)*
z0%Bx->UT;7y=J*#uz=q{(cE4@1vRR88ZQuZ@2eDC&TUZrUAJ8C)>%F3<4K1P5OkBV
z{TafJ$|+o>gJR97+k-amBL{ICOUrc->YQ=_IR{0}6l*XzXHfu018acv4&?Z*KT?8T
z%^GLR;KpZU+@AK5s04pJZo%y+AtXVGyDYydNFx|*JL$ap9nYktwQ;H6{&Irh(a-mV
z&wi`kfOQcj{l<YaCy|gKpZlGgeul+kcNz>{%mSVg;~UHAs79zVub$;IKXzB@O!&k5
zUarS)ySH`kPHCJn>8-ZqY!yC8I5G7L&*^G)(PlOp9C`S%Scq_Tel_)|WN)~9Wiw}@
zs>1y|<O$Z7;Z$}l950(xl=5#IECT6>e!jFsHjOeJD0B>WhXd`}Ws=9?)8_*|$ui?8
zEH*XS$bgodj)=mZ+w_D!s{MF1qyNM0!b-ZDXvh(KRSrro;u+4CpbUA(!lUzWR>Q1I
zvX&?MQIu(OLjlZlN)39y#ar{FEFV!vBK7fFr<9)`zz*+9DZ5@gn_+1(>+t6g{;B}R
zCg;@CQR3%k2sjPCs$nXI0`(nq^sDc3e?}!T*k)rAQ<U_x-9wbo#n@S!UbzJRUg-iA
zr3+)OxYs@yeFHAM6&#cNQM_9VFMsGm=FC@HuF^Uv)TnY5VSZ5-cqZ~aP|@3lMnK~8
zp8+i`n#}YA4tDd5tY`|oO6}(bT?979TrhtKwP~hJbxi!ONxV){w|VHi_h>}Ylbz0K
zzoAS3T{L4*CK(WdCt*LuTDj|fo?ywE<oiAUj%Ezf`ju4otG1DQ5MDHa7$#_n1W-oh
zN;%+!iW8`^mZQd{3{h8UFzO!I04}5YlEAxbpB1mv!KerX@&pWC>v~{M<h0Uy<paIT
zw(foolue{zo$b+T2aQ*a^zn4*6rIm9V~hFtcUi5WOZ^7~+7+9c1m?o`18lkP9e+*g
zfxkEc)#{ewCY&MYWkkm9)qx6y84H(5vt}XA*)asDzf+Em<{iC9mvh8);=g%toK|U%
z&s-9g)gSA&2zJih?Zh4rOZxFSS|c~HGI_E{H-x8<H?Z8SEUt*ueo#%y=bNw0yH;;%
zO@@p~VFDe@t_S*xWmbe$Ew&2Z;QRsV_hW-z%J^yQRaIY}niA41DyZ2z{^K)0K2^vv
z<VY30Oq3$gTeNu<oYc+JcvfQk^~>Mwd>;I4*VX&8Gr?pzbsp{KIZJY5h_ze;I?DAz
zF8^A;;<O$bDPxGjU-IZQ_22v*|BL>kYVy~JQj?b2!PC1d5=9Fl4g$-l7#h)@gu>yY
zCKwOsvi`my0C5hKz2x%@qXvi4!?ZLCZle+%D7s<JEb*UqV!jg?D=X{1kA?~m*fGkd
z$B?f3S|S|qsrmEGN9y4Ip{EOBr3au$MWtbvdri3pnax-1)$@Oo$VaeTzj442tI0j#
z^xtf3pZ6R|?h<TPft}7i#Mg&4ES#$y?6z}<)aQ1Sb))8-+xu>*B4A`!#lmi0)Z9v~
z<;m})xRK%sa&K1$C8gN#X}x-?_0x;_2_-W5F(vck5K1#&Sr!EHqn?ccilZz@zsp9V
zJ*NQaHB+X$D8`{v^6BXm$hBpQ0`WC5@l7}}@(&64o25Z@q?*fOy%M;XWq6FDr>cPP
z8J{_ZmMS=mo7=42Ksn=ZsYwGIh>cTmGG+OME5dwtJTC#yb~LM0%gT~k=>2ys-VaJB
zd};K(M{@i{Ve}NpxWR#??bT!CBr~^|#PvX58PrFNNpl7r1$%X}&K7Z%xsJUDnf(~9
z9%1obd!V8H3SsfGJ?w`h{y}Ez%SWC(%qMy;-obcg4HsuT@tIYWgmyC=|6ji0XAD&y
z5=b+_#c7FUUjJJkKX>YW+{yOY2PY^=rq674hdRW*C)T|McgDx?R&7#n8Js2(K-1=%
zTtNkCsKs_a5BBXy#MQHDxk!{cx5@S02{fGP*s~fS1C_>2sIxe5Qo>351u;hk$|(Pz
zM~;>3U7ERp*M{sNS0$)ks>^(E&R#zyDCr-~1_xcG_U#PEOPS60JToATY;l{a>Tn_&
z);)`35-jA{gS91`f!vj%Rj_cFymN@aBfGfL8$)d^E#+hmmKnFS^E{Ww&fu_*d6d3i
zxJ0vS3E!lOB1wxrTEo&2+Ku^7+^C%nOlt5C)YQ27Gx<>Xdu@GsRpsx1hwURheI+l!
z{+XF?`B4MUD(3KCIWRlG`n};dpf`SO4tZ%p^ku5n@&G0b0~n!(<ryUCiu1z9vG_*4
z6fOZG+Tb?8J1AJZ>0?-QKeH<5Hf~JXe=S{VeOrE&Q`q0?b#S?;AM^_zLL&WbrM#+j
zAE;KsUpDb;J^S6a2J_ZiJa+0oN&yfduM>0!T+<y)L*b*Yo1jYK)Zt-F{W}+XnZ4V^
zp+c$q5cAP2vD+yvUtuheLjDw}ST#NW+Jk33?s%}m_{D?bE=eX$ATHdTlOr&rLC_!3
ziRD^9P9Wbx-T(0}CTEmaRlMUYSW$<ud|~<hG4k}8<UUDvq`6GOu_u8c(GT{c<9Y1X
zPbq|Ns!6kZQ<z9!96-+5K^|S-Xxpl`OxFKatO>}w;iW@?a{nAt1yZ*O&?7gCwXR^&
zs9=OjPw7q_uAVdq;SDNW!fG_y8ujJM*<Vhrc3SSgK@1m8s)?>JyP7$G`&+&DCJVJD
z$t(V8fSGegJw=8nueJcBY&1tI3cLJT|5eo#F*Y*fSqij=fQ9aIZwi-vd*lyqCtNNv
zeX|=R3ovzEZ&gB^kf%r&&c|Tn*)Imyj?**yXfl-x^BD{p+v2$v(LZ(QonET$+J^I^
zvw3=>gMijs6QfVpM@uTE<CVsu$wtpn_>9d_*gayKj@iC_jnNNeVg9|k!>?;KpOMBr
z8SXE~o3Rm&P*~|KNRuYQrulHD@0$=kK1{R)`9>dLjU*?%o5szLlat^PDey_BW#<i<
z3z>+GGD(_%s9)+d{~J#|!C1Hei==q;%Rnccp<_UC%uGzI60NYa*#2#PFQ&kinw4P`
z5vDSgd8IXQIcGDJqLj}Gpqoh)$ANwz`WiriVv^h2%gukj(3qbZIBrjYG_qa=TcuV2
z5SES|h^6#H%gT0}AqsZe1W|BITpX`yYdyykGC_ywm$c1`Eo+e1kd5Fgwol=Wwn0-b
zTUMeX)EG1lFHkKu8&d+&5EVM<00S_bATdOrkK@Ppd-Bi0j2Tf|czllxzHU_*y>#{G
zbHdT%M=o7ilKTqj`#$#mGiZA^|29%rwRZR^xAVTZype~OCZRFD!ZL`l`G@oeCv$UK
zp3sn<&9B9lw&UOHorrOkZsi{-TVC@OiqHL<V|^}ZV%AoCU|t#8M(&NM*NHe)KV*13
z0XV~JQ2lyx<gWMp&ks#PI&q!ksEEBe2!qxoco6_E;BJ9qfR7X}zxKDZe9^Al_--Nq
zEldI=&D&k3fmQ$-x^dkazaEpkh~|h-Y)5dWiMr;X2<L#MoKbb3ujj%DDECj)R(v;N
z9A_)yJ=KhquMu!s*bi)bd;1e8x^o)SH{!d%=cj4Vx0>hig?vBXnlk=PjH7_`a>Yyf
z2`FS+w}Q~6CRb53*DT>y5mfs8H`BH&eR@Gr(f$m_>{o-K7jB31H3^vY_F~<i$%mv%
zUu?oaBR*2O0q07{x@R39Z}eE)6YRj)BN@WlVV^m@8f@U4<3S{RivS}ljecw{ErWtv
z=`zAinX>efDGT^+Idu&HO3KU(s3<X=x&(u%yiofo*5p*OzBAbnoRxr!k+S!vD$lp)
z-l~DI0oLZ-YHDk13y2iD5ou|F6s0>^^9%_{F<cR+Di_y95%;X8uHd^Rpc?<S%egp#
z?B1eLCunB<kcJHu@=iA*Yx5WA?6&JF+ja_66ELUnknaz`D?cDjgt`WLXv^|tzYxCj
zVG{mVgCZqHHmLjo^ceuG-+Uoqh)tf#k@{%v1(4f=@jR?(Sui8+Jx~hjgQR=T^-r6i
zNN2?P*7=5}0Tm!g4YspKVlM-vsE~&bOm_<CEa|V@@=>2OyuGgh)vKm@ih({w6cnEf
z24nlOKX0K%oX2q@&h)E2_a;Heef_dsY*jEejfhh-@_I?v>>B{PngQNQK*`g$hw+ln
zee*Z^en;jOt5?p+sluTL&~4>40JXN~fyVw)6GU(qeP}1sJ4un&$dUB&T@PYOb$ciC
z(p2x&i<FkXY8LIgwjblQHY&P;cur&Y-DVgB;Bx@#0ra9dDrkpv+5e#O_Cou~*<61V
zrLmO<@&P+zE((7bg2qpQeg=+W)dv>HafhjI>Xn@JM5-g>FH<{OYSK^k_#Z8_rSAPi
zJ}$lQYx#q56F>$`=G2$=`ZM-ZtsdHgu|EfKT~%s)%>?+0N02`Gs6J3Ai?cJA{dTgq
z!6!hapd8DDva+!qKxl#il6M9OI2>cLDCE_0iwD5aOE~EI_8=IECV6sUnqXXOtyWKf
zJqoh@dvxm&7dvCUY^#6=(YQ3r${!P$Z+Xs{`pi@qdH_(TtRBk)*$QklorHJ$#i+!2
zxJZU0z|{bGGaHDjvJDIzF|Yk;N>IRrNj1LDvYSF8t6GJOzaTo7IBG=`bqfdpN>ut_
z>}|HVo@H8`BIrKIPyns={&2p_^(+*3$`^SxSB}qkuWQiZ?*;hlm5;0HI3_FE%3dza
z9gx!hb5uORhlhvAyBnr;fI)WxF;WSL5-_MtegIrX9(Vf=a(<`w9-l|T_tabn6Nm6w
zH6VOx%G6jV#K+eF=$`S-m+urmKg9!S<CoV&bOPWUrWDqIbKW+EagS4TG1!Q>Ebkg^
z1J_?x2O!)fGeFqh?uYaI>=Cn<;73`n+~n9=W2Z2!y57GtDBBco+QsYlzdnq31W*dK
zrAB9qqgQHXtcm{`i;r0hP2dC+cXBD9I<n*KLiAbx-E6JL;~<FYE=1=N2<C$4-=Svd
z<Dk;r2tW;!eGTpq$b)K4kRg-5eg1^<xA`CMz<z+d8VvpS!+*=y*49o=O?3o#)W3iI
zua?gI8i<rC)3(17F$4s)=>1Y@SjRezo_xyk?^>a704Eixrcc(}PXWTD_wjNIfLYZt
zgs5J9SX+JbOQ3EUd{z3N4=nE4fEe1{XF9?uWp-c_J-ND<AU!fIb8A~$%LSt*mzBM#
z;voiTadEM$0kkS8QLjnHxcHjVO@f#<H9?-_?`@Nlg4N89KsQBTlprc>3mG+t5Di$@
z5c^(>$B3{3_Kb)|Bum1l>IxU*K-8HNp||45Q06^UM^x@bTjzy7iw5{N-plsCTT{vh
zKZtmTBImtJmEHlJo0ub8cg4^>JNKOEW0?*`^5>zOAvIRNuJ>m+xw+MjK+ib^F94OR
zbfsWKhK8{HIzB@*wOqTUx8z9Pkvj@c@xP4;_w;*V&IxFFGI*~JWwV84bQ<t;cMd5d
zc>=n3>at1g*pv~NKjGzJ@MLDD22{11-3hDx=~4$7HseO8Sm)Kv%~I_HZ(7tVZ@5?7
zZ2NKPB103(f9m*0*+WKVun93(O=ca0&IA7uL}8D6pjsf-L09kHoe%{Sad5lNb<n-e
zYZbF|SLe+!;u~~q=r_KCf&$O|X?wr}fi>EEPO7wvM@tjgs<Sh4Y8(!_<BSQ>TfH;j
z8|$MlQf&0ne-bT2ze#GZp)&Z!py!ocFTC7N8&7W_MT9(=lQ%&iCoxc%R%ZrD(Wq~~
z%t=d^L@hziHE%_g!xK54_~GH<2Us}KRDy4hcC07z<tvT)=_S%PG>k=0q!<0iFJV$k
zYej4a0n_hKZ#Wyv3FUqlYFDFL!LQd`PAA<_g}wpOAJ+xhSB45ap}d8^kOMViw;a-I
z<(jN{+adP7pbOq0v5m}gDkd|6X=U|uhLt}uwJTuXO!VXW8L)XB7|n|AR6|vK?hFqQ
z=RZ@`v?C)6jEHy?@qpQ}zeufC_BVM?mY9d7l~o+ASpGHdr>I8#INp6>XY_sXiY-r*
zn&$y(L(N~~7Pdo!TJLs1m)vbt6dG$!3T|v{0K6MW0&74??5>-uG)~RQsp9a(3W9nc
z8BNmxu0Edk#%?6T3lPwf6%^kUyhaLdw9R4Wn&U511+hE#ZR`cA*qUfwSByqn<=*IW
zQE6@^>;*EYooEYKZK{+c3-1_%`n6<}uXR6RiqPs9qJuFTgdQf<xLM`7xovzubK79+
zQ~Mef^=5XsMeqr4!81mEJV3B}59FJZ)sC9HgnLx|^z<XQL`P8vDM!?x40QHB<=o4z
zYQE^4u<=l0wjDhUgU-V#Z5*fHi{{(K(T`*bhqMqgNZ$iJh7u4I1z<jB5gPg=S}}L)
zl~%8Jl-vOTVggFT!DQi33b(P&Be_Ug?0v{idCCCwMJ@Q4b0LI&aMzfl!!@G5^QIV^
zUIl&w-qE;Z@w$KYA+=d<Z&STc`!-1txlKMjcpp>`xee)bkk#cN*R*kx+`T)LMaN?X
zy7zz-e)Y#hMA1mtuQz{ZRn2%R>PK43s+J>}jZ&_X1oYD7TALyDZ1Is?8HS~Dy-I2^
z%Uzn2xz>wh)^H8@tc#4nKBUqmp!N8&KT|g>>rgtNRqJp~6lGivhIRsIw})CtLV{)p
z2K2QoY?gs656IEFLPG59>_Gp`akX8{A&<xC_wV1p8Df+%JsBx;4}HpwKMe{bKfciI
zkSTQCOUk(^x@;CLyu)HQ%~jkp7Ad?#qTZadM;|72j|7kp6?yU`O7HTcyZ3_YD^$U?
zW4QsdU{o|@@r{yK?ipLw=`V(xyIAb<r?jVYt{dwqT!snT8Y(x%c5uPk`+egOE4H|7
z5$6miGI#kJpa^w`;EK8|H~U}h1LbdTs_oZl&6JC8yZ1&8S$x|;S422^0rlmV4BDmp
zExT(%cCy&}*HGyW|Gr<p$_og(Dgm_eD)5%Df#GvSMFo|h?f07)nxpGKBfyy+EH)ef
zj8KP^QU2@atGP&<fZSyDBV&;U*V5@u64oYP$-LVTgLX^C1K342$L84Zt%b?<r@8wG
zT;0#IU7&-G67SGxYjo8N3d?w=>KjEstF1M>iTj3Ofl{^gwO%BcgtY08A4Ds({vY%;
zIV$nvw9BW%*DvTNuouMwg_?utI-EwRlU@1z&aDqnr+Wnja01s6Qg=U~{ka4nSHQqV
zrv*&?p}|4a_p|Y(!5Jwpkw-0UBssqK&}`S2#?JQ@Gg@YKj(4*_Jql}$k_ftoS7nPA
z<d6uFMh;(2kJ2s1QVCKI=_OlZOgWE#51#UW#E$v21X@JwuvV_cT6yxua08Vs)}XK^
zH63XqaGR{gaFql`7Nb-*I<g3F?3@B(6h=h@WRE`4sU%XjCut($;T&<S4QTx?&F*fZ
zN4cp(8Pa^66GNSYHi0wr>gXbm>UtT+Uy*&cK&Fy8EWcLN+7PW1DFu@anMp~N9uq_)
zGbmMD`?`+};ynuP+qg>{AMIOJx&mG52JzK+`@s*Io12AQP~!IIn^ctd!eU}nW)dPi
zE{A>1dmgqn4;6%cj+X~uYAKlJ`###DuG&7Ev9vPMHv7{B1ySJvm%9^(hr_0ASiIE4
z;?h}iil~L^jd9^ZjwXg@?ed{2$JY1S3_dN5p2>`ED$HB(g#x9ES*U*T$BX$CX_%o>
zy#dtrY6x9*b<buBD_b~;p$^RUud{{fb6f>nRvIn&9O`(ve`Zs;yFPjo81eO+oLcZL
zrzGdbT|BDrLtFi@{063LgC19+)RsItVFwGkUVqMj`sZqySU%t?d{j8V?8v(4*PO4=
zpSsD25i^wvw2+Ymtij=VAndm5*LRNO7|2yUWG}N*nWTFf$#(OCt>*YbQN!gbF6pQ1
zFMW28&V%CM^GCk|g@H}fc~Rqn2&E>u<w#P!E3<!&gg18Kl}m1xUw#M}agV40C*Sp!
zuiSgxZ=OvRm1~B@0C|&DBU93Ee;UZ1c7ZZ95t$;+kx@~Jsi!;I+Q0hxM{?*Camz(r
z-|{a|Y)Vx!XyA12fzPdFVK{94!PE7Z(7@G`n=w&wFTE8T#F^^)K?Cmn_Sx9SxE$E`
zoj!q+#+Vk!I#Hai>vX=eT>IUeg=mE-a#_Yv>RvtboWYy>M%IopaDM9y?BlyOlOf^V
z8vYdS7*0%QJN$c#Rgf_(9`kxG%9`idPPsQj(j>ZOEsiPWf;V7Q2Em<7hY4!Htr<@`
zN;<iPb4Y(fAV0{ca-LN{1~@zSZ|($1k$0lEWl5)js8BSV2ke~!cGuuas{2xZS)BXp
zt&xJSgZKBrS}Ig;a2!MNWqH`;bZOu{yEGbg)oG*wFOyo+P#czR=6=;9fO3vPDiz<i
z&Eg=vp6cbd-5jrUfYHZTlwb}+7=tvxadw-V0cKkS{1g_BWQdXwENs}t<$jfU08W<Q
z&wCU8voD59g8j7nCEv&6tme{S5=O^9&{(z%QIymU+)YC6MbEld8{}hwZ82kT1Cw5P
zT`GWAMnw5pIQy6N|JcRN`a>r4L24%A_OtifBN@jF8}yHh)G{Zl)XKHlGyt+`V{I)S
zb{Pr}^>#K=&vrlCju`-|@n2fzFvHQHq4;E7;P#-Y+ja4wXXnfh)SgB$37G4s{Fkmb
z-g>!=|8WoU51hiaU!Rlm!{Bqn@=gWq1;--9P36y4!N3m$N($<0ZBjMhi(HLUU9Rj&
z1XFTAFx+U1)FD}0A_#m-!ZZ;Jo`4^A-ChN6@{sGiJ-tvPTmKCE{eWy^y-58Xl`J@$
z>T=FcWhW*_NESMil$IpK;nMS2o<P$_XSt*|oP^2kPb3RjXA}+Py(P)(6^3o}^CdZD
zWqf);!GzfC>sutQoLsP@D-dOZcB8Zs-dsG5!j7}dtT<*}g*MhHcg+XcyuwvW9_vgD
zo~_s2Wh}5*<a2MhV6wl7ey3UPqbRK8IsqViikNqO^3ClCU1_nh3n{8;398_gkZ}BY
zX+L7z=dxl8CU_|=-3fvQ0gUVNoY7hipDuL_L>YUb`*N>b^99^s-(*RPJy5xjh#gY<
z=9}h*>=nw<I|ZVZUa$=&a>|w32Jvx}PSCvP_!w}rhWw*NHFSwvgh{+#PSad|SdMIl
zxR+l3iMcNhO5y^$=R@50>Sc+hc@$$wmR}s3a{Gdt|1l2?c+esv>7u+W^e~!gs(7pg
z=%j=WC=xDjWWZPI2ef%^dD4WO2kWVVpFI_oayeRTm`@WwXqze5@BI-+Pja7syBpyq
z;eCwvl!bYcnS~{WO0Y;Nk<}C+k}GXKfbSX$xy}~5LS2g?zKxb~G&xn27n)ZfNrYcm
zBdlW4eXVbyuj25qE@DT<MdUK$PR8+Oh7hE6jyv+@eh<Ns-Rim0W_N?Q*hx%Gv<O$B
z;I1%)B8cMj$|kon0~I_F>0eFn;!^zE1{@h?L5-gL`}x@Q9)Q#ih!Ox)3Ai~fQJ|6I
zAxND8l_%kGoQ9ANRH_u*uqO)(oQBQuJv=6vBe-?c^Isd8uyrjabUZ4*tB0IpJ!+sA
zpWDJhcM_qHGy%!VHm-1K1Ewy-;u!0UKQ)cvg#w!n2N~a$(N=9ITKse*S6cxn-Kv?B
zAZ46^OI?oi@mttJc$0;%ApqtCsW0dFhf;X;?{LW2G?H$<0icS{F~{FZH%;1cyKf*k
z)2gm(S#17O9leZKz~rS8r(OkKmE*7H%MXW0NtGriCp!VP1BZxuI8&6Q>Pr$pkh3Kk
zg`=IpYUzW8Fc)3uTmfa-6=KMkdz5*z8B=?5(y`(0dS5^{w`3l<|H~0}aWi`)?(IBV
z|7bSjbWg;2QOIH5eC-=VCKh>{F{$esl<)n^qAnp}1@?R11HKvP`V(~?(K=&%?J`+n
z0Z?!Wpy-uq7XwB92YV1A`Kbj@E|do@avJiL9H0iiUHYY$EqaBM;Zg@OJ{E>lK&t$e
zV2#Xp1SUd7YWEAA&EzT~TMK3WSg*zCK5myegbB!0(=q4CBH%`2rP(nueRDjN6?d<H
z!edd|;aIMN>Qn{pSPzZVW05}BHO|2|KB+NU1N*4chjU}1dpDpz*@8Q!_1=7e*OxVk
zSgai`NdHY+Tf4Xzny8{6gHx<Bp!MR4Nh*l2lVv`yFm58=Zhaq*437iYR_RhT!bi}y
z$7QE-)lY4~p|Z2x7pkDBHssspjqX~iA<<>vx6*Y`9-J&x!6kh5Nq%R(#tQJ_%=Gy9
zocI=ipy`)g-saZo4x#Ffru#F;Sc=PTWm8<mKEPncFq6*X!(<77vjaWUxPBp{X8?xn
z2_rx>=xf!tM5Dw<_5TFDGF-c1@n6Y3ySmHkOteZcxahV%o2yVwgCQK^ta-C{HAWEI
zYHO$u1JOj3Na4=wsZMc?Qo3&mSyFlWa>=A;x`VfY;WhJWG*z&Qu&5&Q;LO>*DIogq
z1X7I;^lwgf`qXlSxZ!jm&A^cbrZVZMWr@3;Ujo^3x>UP*GW+g2h9D@z9-t6pY-&c;
z=CaE|H{i!@I+m?G-y@@^l}2z+)w#^>t`2qru2Qm-g}sM&tLP-G3vdrl5ra}E{v8YU
zAmXN|%(m<89qONwD8FGWU1>)t0hGy_;Df4yBD(;=z%40=@SS~xva9Y%hx<j45~$vG
z4lNB^yWZJ{tZtv1Uca|O8nemE>+OpZ3~gP^$!~m?@it&BwA-7`xw*S3;cW_KJG~>l
znz``^h=HBX1jsddRJ(?p)J+=FY?qosW%e$1$Q-1~y?MTgl{|8q2M44=274cknCX=&
z!uaNH3<oO2LhcosEJ6)cxWRNyJtH7C)f_%4I5Nx7!!{=K8QhTeL7ww{Qu7IA$%lxm
zgL*?C(-Xg{;q4Z6ctN@zWRSGctLFC;e_SUoMZ6U8oSVB6Xmi1~0vd`JI-L(P8D?CT
z?am`d?i9vMZ!*~&eC}@715>G@L}2Hh9WcqIwf*9KNz=j`SNig+xmHjP9_pF=OZ8v5
z2LevlM`P1%fLfuu6tgem(jB8Q$9?Crbd}m`^5BK}jQ$r*9U;&DyMxc~KFGoyZDF_q
zcPZy5H<#GWT_<uvnz;<>R5MS8V8g=u`gFB%$Ngm@)eWvhgp*K2zCFt<Cod{P?1&Tm
z#Xt4sMP_5kv5eQNypfNc0Jv`AA}A;-WES$r!n$BXr6GU74|dJZFiz(HT&DM>L#91&
z?s~$_&A6!ymT1&qJ5ti@c1KdGe`&=QOGictKY8M0XJa>jet0-f19x#j)nE$gmK3Tx
zNaLZ=kT&lV|Ft6{sDJ(_QCuLZSk&0$>O;&&^sK|t^_x<As&bK|w!7q_oBQ$=DfIim
zxuqJm03~Ofv1oR~y{TEIvBJ%B=bDvpjlfw8Pt|k|{=)M6>;AVFvu~vmLO=i2!}L3$
zu>hXz&Q)oLL6@fDZdhmiU&^C{v>iU5)Zqq_4;IBmt5MDN=+`)1hnfXn1|9^ED{nT-
zm=p)Bp-%I-I|o?;h(M#%7xL%TDUW98FLk+Tet%)m=Lp$Pm*QX%^&8O7)6t4r`;{<*
z{NomU^!#-09pf>HMwTR^?K{-6TikZN@15I^zoLWVs${eM@^sXY;ls8wH!@=MGk6N|
zGbw&zB34%UZM&zH7;5AWrHL6NAJ~C30A6m6qO2Y`w;7@qb#DAHlfsyo(xKHc#$aYo
zePP~^33?9?2y<!b7(Y!(rTImi;kKgP5U#47yL})NbOxa>MdWVx`W&u;Z!g)AL_$#&
z6^9haqQ0b3|C3B^)Xu;n*uTRM?M5IN1A8ImcfJX9pa2Ph_Gs*K0+7`9B5&q&JuuOt
z`ZcO+yykb(YcjPuHlh9cM-c@$ltIu(8ryzvu)t+xaQ!wdfI`(#;>=~-iQIBXeY39s
zkx_R(g$HO){zV&CrtrzilR7ObGJoaZO^eFFnZbwmoyw0u9d?*tdW`9vO!s95;wwav
zTjH-kQPY*L67HLLsydzQ5`Ws7wTpRBVOO^?G#f=fvk8Ja$W=2XD2WEW{~sSh0W+!o
z02#;v*JWe{bbf5&y}dq4Os0aLjb5<Z8(*W~Yq;4#=kF`i!X54}^Y%q0=hl7#nvh0|
zI2qUgy|MT3>3mDzK{<mc(x!CqBxtt4s!asIc{1L^D4QDXM?0ML&rngM<c{979-E=A
ztNYn;sNxMToAM6Lyro*TySUs}{GPkt^Z<SA=g&dVZ3PG_?Vo%21JG!vl8+(5aR$Tm
zf65Vxx2bXPrs#JJ$dQ;s;ksIpZQmvV$GO32*F~GyyB_dq54gLC%@&e#tkDV2D;{?C
zsB3)K`D{Z&Wpwp2aJz&!{fC6@i3v5_i%s5p`}SlEydvUpn~Dnk-tzfU2fet5Du`mR
z|LN%6m*lZyDt-A?e~V2cS4X0o!F&-=bANKvRB9*!m_Hqyp8nxDOHgr(iu1L0n9+X0
zHc9(5Ae|t9vD*<)(`b*N>vh)RRVKrH?goUHbr|+xhDq4PzX}BxbL8})rqga*R2vid
z%Kq+bwwNZs1fl>8UIIKw@*`^Bi$-K5M~Z-cp~6t`!UsV@D|5H0U}}Xrw~ylGU5!#!
z8heIV8ARWxhU7}?9PEKb-><Z8Wi8Z+x~{f+A1wl-aAhtltw3iWfnhiH)n75Zbsb4w
zz+!j2loV(gB_sID?zT5R`FLy);S{RC%7)M$-D7aaH}E>ez2zk>@%j_^qYJu{eNoKZ
zt)(0JEjc|<dLWk?9=j&5$zOaH-pc*OlaqjV*c6m$++VSBZ;t)GOj9L$`aZba5s>HU
zLIQ49e1=ka(SO1ja7d_6a|elBxo!}<3zR2)v|{AwfCg!R=B(V^%tq{JnN`iU^+6Y%
z=L4;;Qn1LWui;~MUV4=#IT>ZgZQp@$=&@L8VUE+2%5G42T9f4+1IX3;MkR)A43?H-
z0lJRTs{$Wv;B?>w9!Pw2W693QXtnIePrC#p%@=U4@qaaK&WM+KpNjAsrj=qID#X!~
z^8#~uwt&)wOSlX5T_`J-+4DZAyCR_@x#u5E54`|GL{AsP%k=kOZMPQh>%EuR|H9@#
zl1}Qoa}}UvnW}zNq?Udt)D;>U%46Ev7ega#HIO2a>-Yf;GlqEFm3(;Xe_Ndw^3n4k
zx=1HM*2KgF8182BrBrzHp9X?sP4B-Y?a2#4!3$s<?k}`!a?K5>evjBjLVZhaKN3O3
zGkrTPhaeCRQ$!R5Dnp)oX;Mhf&%z;fX|c0mhw0tyIq-suKaSoD{U;T10fryg#elEi
zJ8|tWw}W5;*NRPMfYtyFlPC&o@W_03oD6Cnn8ajJaYt-aNdC8z@xKk3U|ND5Ecwkz
znH}^ybU^&79X6yB1wZGW)A2jY4+oNej3WxT4;qV~BB!}}Prt+=-G`p4G;Z4_84I|U
z3(Jup4!viJoGR7;G+-asue^DMfWE#vF54+VZ}{Ji2)DPvgA9&oEjHnd0y|qI;>6h!
zb-ou~sZdmeD?R%yC*H950G@Wo8MM5eSyMN+jBe87zX>vGb35>bQI%TOxc5g6>dL_C
zrA(S%d2FkO;1FGk_HdJuku5bi7NZHoKL59`V{};kZ$<oPU~o`CMoH+ya=s%2<f=}=
zq0_7m!pna;(8>KVcGaNs_yoOpUaT>LFibDgKsF6kM_gnGyEeL*u7Pv8#%G&gK~u=6
zDt4FLlDRAya3|2#V){Zd<bKx-8%*Xrm?>`xyz_2g5c@xGkF-m#$!UR<l2X)ZzWUvd
zx7?skE_Qae&EXlDHgp|Wm|Ecd<Yc*uzRM)%sVmy6P3@~+_3eq7GqrZ>`3l&z18uN2
zrSYbddj7}9xjH+nB~}`d3dmf^Ky6_Q)T`BUA>pT5Cd7~iz92wWzfh=>3K&*Y<d4D$
zw);OVfdNDi=s1<SZT!rZ@S%V8!5q-UGew=3>jTFQ2Is(q?hwJV;6l4FJrE4;*?}UX
zSh)}AO(RXtsa(E@F+v7i=p1=TV95y0R?NZK&_<Q|kX1lA{;bQy!q4w^wjS0%)3<uA
zlYDFXla_Gt!%Km>+PA7^?mGmzP=x(Pe-y=_sN2LS(B0-w1B7y5{nyHGfiLF#?b1cj
z{Tq!V41nTnP%UCVX+2wEXgB(B^^KZ}igA;zS7J)`V8_yjIn)m=Cv=>h3%J8ib)gK(
zrC;pc_O3WU>0@kG>ro3%@?zX=&`c~%PQ}IBkxt|t$@{s+qqXi3h3ID!K<d!rHYTJM
zxlK%gncNsueuPCH@$ZLQ*}0=)l3}?S@nUv}Mzj#!?G$cV`>7$sR_blv&&J)Ea!R9T
z0?emx)aO@QJc>E!?ElZL1y>j~lmSf-fFo+&QNHpu2OsEh+l>(3m;Rg${Iyysv%ARV
z_$#UM`ZriSN97%KWU^u`$&sPb%M{T{(m8c~8c|2Jw+C=KK&_ci1qF#YFV=7TioHWF
z7%nVi-k{)dtspZrfZbN#qmH7J?LyE~Iq~TY0T-Q6@ahR=At(AQW}9r=RZ(3{0hk|;
z(?}P<=I*kbt1!f>OYO{m@ZYax^b`M2VG04*YCMwyG3~2y#DSt>L?#~~P+hM;a{@24
zy<A@E-^4aqLHsLbH5e*pH+rDcbYfKCi>Md~hp{sXm?uhuaW7x)wmF0K&g=B052U@X
zGsHpB&_5H@P13q;_PGP^705XZ{%(?IKjHxR{)QD{nO<d?0O9rVN_lNabIJof{`v|5
z9_x<#^M14k`S>(29KyW>40*q$vfC)xs(8j?P3Wt0$M{W&;A%yd;D5}<svvIjI-uA`
zgPNRSs2zPrGt;6)ni+qMv?f&J4t3DHC0v3C7rj40-A0TZr1Vj9IiZ}s%5CI_u;`Iq
zQ+#j8|7!0m!=momeFX$*1PMurL1I8U1tkQAZUzQ$$e|5NL`i8Ch8pRIa!3cHlmSFa
zC8bjgdMH6cCC0PR_j&i;@7^EwKIb}L&UL<UF^gfX^<Qzv@BZEQU(PGGpw>wbo+x4|
zvh<PKIBb0Wo$mLW{j9XVbz;pC%SVd8OfIWeSI`-+d}yq7RT@B?dgYEtJYO!a@6&=M
zfY35&@5hy@(R6_TvmJGSPUV>w5Pt={-?XnfZbHqp$)uQ%cm=ZG329UY0okkm>+9U_
z6)FdPJHLsaxe@k!Tl3dO&`9mxm9HvRp~3vL+skGYhIZeU6`aF%EONj7<=?WCj#5ak
zFfZzz<UXTG+`LwI5|bWbS*XIk$b7A|b@aq%o3kX@I&XPZ0SosCBTs?8JgATpt*9z>
zN>rlVop6GlObmru0>{XsWpES_Y=s-{*48cfU1rEV+H;osbW{1-@O<dxR<_WU)}PTa
z)<c_Df(nh6rCNha58l}8pab^TrvOUy^+V$hpfVgh8v%!<YwgaHs>zL#r_ihW>dQ9J
z21mZ8!@+J*pVr3BWRBL5a-$}7nYq-_PpJHHaFhny0n(F${Be?%2W$<qBTL%%d%E=R
zM<~dgN4!kVmF0O!F{7o;u4CG<Hp<6<Z}*J+!&pCZ@BEf<J`O@XT<?IjeP$#6d8ksD
z@Y+A*!0UNBaQK{oOaROSUt(~m8Bn{Ia&;2VXJuxZet$?cp-TL@N^`ZJ;m-YUsS-|l
z7VDDyTTc=XlkyJ!dg~>(Ds>yg=yH_<JckOSkM6C#mS|PEwe@*Zv!uL^c<_S3@^Hkf
zpTZCkEWbU6#Fn_ZT>8}fE4@5~xcr49>8Rp(SJ3wR=GDR0urLpjIuO5~s&O?v@KZif
zn<daLq^k1io~Ymgy4IoI50YJ{pDU!1pd3qD=Z2&kw}{jW!o{p;ySmC3PhC|JbxZ5a
zGkEn`xrX|YY|ZqdcLDpS|32SI;Oc?^KOTg)mYk$y1VY*uJf40mmvh+tklFjfG5)9S
z!!M%mzh2#%C)oaV^SpV@u)gIzp<ITe^^TI~$X`bV<(8hK$4hIB0BLzUe9r5W5%-K`
z<<Ra@3FpyL!7AOW^I^MDhp`#bH8ohV$scbgvbxfqZFgQ>3tSx}nNUGY0%`)VK7cZd
zg{tWN#gF}MYR^^8X?qHto>Q3i7W-^Ar)p+ki-n@tEk^B|4xkOmr6XjU9Mi~pFTDU&
z+9?*AIz%-^5jS(j_+}2jJxvmR+wO0t^wm_gBN^rqKr3x)T)Mw>!~(@Az0yk@TT?jR
z$ris6QogzOel+QBHDYhT8SectY~8<o$egOm;;Z?7uDl_IK?~M<jQ)X}gVJTs_Bv|0
z)grt{lx}>yW=x1ZaW-ot)H*pmy|?uR?o-WtdJ<>Aikza<_S_wp=f2U#K>OYz9{n6J
zR$RVtqI2vUiPTp$6DUOHG@ay)>SQtlS*U9tdFunpjOff<uDTyBp$d<z#n)G453jUk
zi=V2+Tq+t8`E$Pn+CcOXU@;RfZ_l3q%I<^tz#M?WaVmiqx7XOAVfMIz4GD!t+<kj7
z>#t$8UCtK*e|D2u#u(VdAmpY)4|{byjWS;>JsjAWzxc=x;czrni}WYR-td08Ebwli
zhfC_d`*PyOxevrBx4sOi#~gCXccr)D7vG&wQcliWhhmgR3+6cEHCz30euk?9D~3&v
zt@^nMks2LsZ$bNVZE*{}Mjp#6KqYg}sPd2N%fV-6Zk}}EYAC>Ai*%BDB$y-dvqB#Z
z{9Dp(fg%V*wXNp<a+egwT0BKw+5Xx8UP0`qQTClR2K;6t#fw*ypLevM3ne!F`n3j<
zDF-V(Qd{mZK=Sl;VS0h0QX-2hNpG_iyzt%xGfN>MK(T-L_;Htp)rJoJ@6{F|dS<4^
z6RM_T#VS_|68N|sI;OZk&_r#+!|gptN+A!9epL8Bri1oYKJ8U3Xl1o6OXZMjxS&Yv
z)$wZn<Wef8AEuIPvfr}*{kW*VdF=AxW28kvSM}BAp`-E|OJ_Zca+8t(v%*6K8-+^S
zo!!fFU&C5{YJRN?*}C9kw`-I=eDekVdlyI3{<?X9Z%Rta@N;J!;xDM>OaG@oppq^9
zoxAXno8c~wY%wLhr$kUcA43v;DIHt-s-F*I;UxRHJCU!B^lqzkcO{T@Hu#0s&G)=s
zsp>LWE!K_&IQg{dMLXcV9eifHDJ@-d#Y27_2+H@@Ul~$|x9Auf+g3l<*dg{i+vX{S
zAf7I{j3v0<46|O&RX)DzAvHI@)RA*g{4?&C?9nRW&YJh+OWm<<-TU|N1J4N{8x{Kk
zXHL{NjyH_FnzdKRfe$7%;p3z<r!8EjYRs7(E()t#H?)1S7#(RrHD9Hs?T*NQ<2h1(
zW3#8>dP$*Ng?WhoB!!uN;GV@h6<)>}X{wrB=-<8%S7v7BL9b*JeeO#F0snRULRuc<
z594u?bF4`Fa^LlEj{#%$<DdnH+KZ3j&+N+f=>|Uom2kMb`sKaZRhxAqEA96Z%3*tB
zyc1P(Vu1Au2na(z(!+jlIY?FLbdbdvQCI{T)j3nmDRYb%ic5_0N;S9Yb=^OJqT3_K
zveK9r&iae<5+d~qTUnjKrc)(yUzP_EeE^x9;4~nWvgyvo`sL6sRTnDief2*leF@yW
z8ai(t&=WSjaX;bkyUPLHQMqMyV2RPx1$D%8C#{y?y*@8RW{}X}F<LtGZLxQZNN$v)
zIP@yy9<3M$a_EmfU9P%Iv03;mIjZe%G?7f-K<q5Yu{+1f`LKQoh`}bnk*?Mqx%bKQ
z$mju|jm?zXxzxW7D-V5)Cv2Mpw%}E)CFchAxNAS-6tb><o8XvzwSotn(DQtlf)eAu
z@|65gz~LTr2o9d?@;XcF+!@CRax?Nkx(~?89n|pjTm?+%Ccx|iLbpgq_4w8H)8E&-
zaK5Tj;GgDjw!+ODgZXWMH^acK3=z5V7sJKxq3PP63@Gmxm(IdvoHw6Ix<?*w&+e2F
zkjD1JAF;X4#NBN_7JQhh4<TRuK^6!sfaw6n5pci?05uIvN)J~3Ye$Y5WPF-q4axwl
zauT0;z~Te)l8bhiSN{++C!Atg^2;>DfjL2fi3bQ)RBG<kux&S*=*C*Nz>U$+Zy(Zs
z?O1OB7bYj~?)LJbOl{4p+{V@D1XUdVSMA$RUlk!YZnrnSt~w(mBqVXY`8>@y7#>ho
z*el*wehA(DMCI9*KGY5zt0spw(hOWr9O)mL9_!_Qh^#CIU@(GfFarqvr6CVwN{r=h
zjuFS*uKem)6Q{q>`eP*illauP7kK2Gqs5WfPd=9&)=781N6~Qqr;b8@d;wUBvcoJm
z#L_au6|$T6cbw%eiciMPj%+WZE=?hx6RJ<VhSxLVO$>keee-69W5~(=li>_PyzL-r
zd%0wv4upX4{+lWegM)tEWh1QK!K)pYu0k!#7Hq{2l58_>UE0bff>IFw;^3S)!@%P`
zftv-%bfd4@y64|cgXi0+7+QqxzAU#`Jp7ozEg9<Qi`e|@^rdV6=ET<N7*3u0;lZPV
zAecP68_}Ep9xq@Oj6)Ee+X)N)(4jJWw^Tmp5VP+C9oE<S53v5d=dYGO;cU6?wkXQ1
zovy+abectF<vvi5a^M^Q8{_AFrO;VPsEpsrite3O>-i@P6E6?@S-E5Su7qL-KhXvL
zp+EkH1b76A)ZcxlshE-jY>q6s2Kv%(2Xh%@U}zrRs{ZK)$W_3tBn%b31xBBlnHh+*
z0D`=fNBgAPub)3JT=)FAU%~#?h)vy%g@3VXo?_vL&7*TVOqHaCepGB<E1Us?#Y`~L
z+b(plFNL*_*`1j747P6d>#{dUVjdt_p9$qRNGendV^g~eRftR>v5mc$l9?t(Yt29=
z0@Kl&8uRrz6VI_60W;^=<)0sVTtpq{dSyr8z$@8xS^N7{ndhW+I`KE#82MzljbfNx
z>{-LK0=!hvv}3(`X+%*6D&SF!1Jp(qKxKyL$;nB;Z4FUy-XAjG{1tcIB`&VDy>Bp$
zJu&x^`*aK}QgJwd8dC7dpcChT8LnAmHwq)iuoUQlylmFKFG>ueHSG>|Zv{s9XN2>U
zEV2vNJ^8x08oraM>#9DV)J6ug&EPCAS?*5o>(7QgS|7UgCg$lr+s?C78>UVOU+APk
z>{TUD_sFsBo4u4X@3{McS1F>v?yRfH@R-h!7$VijLwwTnRBocYrCBv&ty~$72KE(}
zW<MKQnde%X;_8%WBe>WcBEEx7LX;p=if>$l>FFU}-ui;((|mRGtrvuzcrFJTxH7Y(
zqz>o2k2^Z7=nR1}O1F?`u=j59R^WoF7hldj!Q-MrD>5<?enh0P`I87gYDL()x82mR
z>~39{1l7aPrOW*}nj$xs63~USyD9Xiy9~m*<fB-tA72_w>~jmf7A9Wv8o9`2E7e6k
z$Rj8pPN&j-9>PdL)*<Ob!E}bDC`HT%hAq&0Fpjg)SD-ACZqe(laWj86Y&n>Aqn{&b
zs%EhYpYW0wU4(I}@p)|CD@NwXIiKv~*Ed|VR&bXM5hs?ZK58^XN?g6K4S4-Safx5C
z=u$G(22FOB;edds0PbzWwx~LW_{GnR$E&0(ludJI=A_)uC84~`^uK0NrwbbDP7Ft@
zP-BksCDI*T2&@q9l?Ul!{^Cvi9-ox0r!WF3I;zjnQ`jeDLmC3Xp)Q;o4rGttg+C+f
zdqDjT`RsaGB%Bv3M$^d*RlAqA5C0JnNCVNp_A83(z|+Jwl+Wg9XVcZxZKzL<+hA$H
zfb^MTLp40rF5N-Awd7J-8&#-YE1F5VfNEva3lFDqVQ|K<`f_CXAnA43U5lNhH^J~j
z&s0_FCm5ZOse^l85AZYmUpam)CfG$gEy<AD=UmB4GY(5v^CrJha<)laU`PX>_-%{A
zhou05an4MgO!*Pv!l%WwE$nJCId+eC<c%1CPuELrxNEtN?8{mI`GK|-`|jb9Xrt%Q
z8U9-CcU&2}adc@?UXHfLS>Tb|9d`SA<tfU}groL`7HxJX=J>C@i2G6T0}W}9br8b#
z!*DOtV7^}jQCi8@`{GA3>ig16u5gHQ*Bp)QR-1@t3+UMQn$#n+nta_+^XSF?N%?GP
za~DXEte=FXdz<pms34)GWZG8b31mdS{T=&<dh{g{ZVmH-1P>1yb=@?!DqjSLS+x&>
z(UXuU7ENFihus;D{;8*t)cKQNb0VcJ<8ih*AB$@#TFEzglQd<%D~5T5fI3`i5LAVT
z`$t{#;!GMQ%}V*%mju`p?;`gSa^}NpwKs{?8Ps+pObjwUZ~R4)FG<)abAu!w8>7y9
z25hp?t2hEWGdj?K=2|umJi`0h)<RaZ-Ahqev=d)$mOyvypvJexx&?@FyKjFJ{a2+8
z)5{9P71NjWEnZVO_8;YL&ap}iM;Pa^oxgttm_y1aUhQ!DoOv1;#zPc?#`!QTYO;(-
zi=UY>o8XF>OnRB-t$$Z#_2l{BGb#R#C_a0YZk5x?Mp9=4H1JbwflnDJ307?H#?E*x
z^)x6d%lkh7&lkc=E&?PQInh(80-)?k!$j)%V>P2}DSN$Q`_9a0R)a85SuyjHyQqP<
zYy~%?tBW^3d(vn^Wor+!O{j;QtKR!Z>7lwBG@SxL(R`-HIhIUCcfuxjn<d-03DY?S
zm#GA6AMMiXP=OSy2L4(BeHeqQzH;vv@~OHvZ!cAi-SL=;HhbQq4?Q(z#l@L|Edr(t
zJR*%eF!cl?!llhu-Dk1$<kBw{yl~{4WVtEn^QMMNM2scZ#{IN!f4jUHI&lMdDUJ9i
zyTSo{-r24$TQ!e&VLvM}p4u2h1^VCNpH)1r507Zg8W3F=LGsDYalABquSp_BR)WTf
zpXLmfa-$~T@iy0MTu$L7;QFZuJfnU=O|i(`z0a#v($kXSP4pCtdzT>OcF3}f=*9B^
zm{1?Sq?DCKHEJ{9yaAI&RnI`bq3e8|kM$<y&THKHbQh0_WU(m>g<-`=J+JpeDs{&r
z%V`C_r#2(y&uKc}xGXH02rO1jN>;e*vgZnRq+SOJ1*AW}!=amAnM7(w+nMlvin5CN
zR~2-yLfqP_apbPxqM?53X80X8=C&&VEJ|E{-tSUOZFHzT{_<;B=eio`8uInNk0UFl
zaoFZ;4VrX!AxH@w$$VTt=?J!z$$7A8ay-v*`Vowp2BjQ@NTGM`XlQJZ@rsX#Z~sw2
z*;tMkkjO|tmtH1c3hMlFO8(n?mDxl)SG*o|;W@&exxq+V2&$ATFey4s1EUY5^1Md*
z=5?J~b(%1#5zZVUWS_$1d=rjGK!A2$YB)|_6h^N;JxcCRGo0_O37fF-EtWpD?2t1*
zv48V1yEM(^W0sVJrD{Lbzh}9^FTqtqNYpa$HPX$Ck89O%f^hiAV<6+U8S=Cqt&9d0
z;W@<9nXGelvW!fckM*E~pU5k(cfRPfB9~#4W-x3W-I}XHkJd94fCELq@<TjxW?3?T
zM+Fc{w!2@@genMg^0$d2%1gdBR`4f~+cnUz9C+G)`~p0(m6du3xx)+4p}zwtkg7qX
zY63Cq%)i5w1@HbzWBE6OR`>d!M33JKeiEs&NP+(gq?^9L@-LR#|L^ASkj?+wYeU7?
z#r9|G|Ko4hUPq?UH+|Wgc+1gx>@EOJ7JXzkVFOU^sIw`^PUM)8yRtmbOySamXDoBT
zpVsCdYx{<=SAvgeRz80`CmPQT0tYAGZvFvft(vQ{YiB+8r@?m*-V(%eo643eACxA8
zg?5i2_vS1+1-)+rUkv2z2sXDO9>N))RBQb%R<`iOZlWquv0|D}iyM^rh*|&>zI{^c
z3w;EB8vW^WrK*z(q|$4Aqd86okjhE7;3r5s)1d5@*oDu5j3&@CxEeVpWzrz6gUI-E
zpfkvw%~yw;<(3uYxx(RZhN7qAr2$%$7#}|cif~Ocv<cFQL=qF82W)<@*@z++xaL?L
zwC$AcPrd`{HTAk176uOw37~xplHb}q<~=FNwmFFGFb5!Lhg!GBm&!;&AO}~ZjFef1
zYH-dNz_SFS7U+-VRCXuqq-Krgx9t|_YIVRY7O24^^ZW9JQpII}GP<5+v~Tsa+;r!0
zsEgb1=4-DTAk^h{XJuxou~w`+0R{@33Q1C<XuJ57RwvpFS-)VxS%ab41iEw=E^hhi
zieCNT49&m|BS|kCnIf}}k6*QV)i&(ZraOzK6>!uJ*lf4Kb@G{Uz7@OKqH>gl#i4rN
zb$+mHXHza)G0P-}3rE=M_~p9BXt)rx`^2bLP`Zz=S>{IHinpiAIz<S!<rsQCysppj
zyt3<qR)j+vSF8u!6_ueK`1&ee=3A)1#M_7ds)(JS#YY)un22c203n~%%|5YOYWcob
z1}4|xLm8Un(x;~683sR4EM@fjIMBadXB^wB%(7glN#2M%saU>{$kHXt^eUj5LfJv7
z_Ut-jUl&9F<P;&r)iFB2!!Fi&eH8(lAYsmk;B@U1<BA)Q)|5mmUa~<I52TkKRCtpq
z2~Ww3zx}{^%f$DUOL>0>L)bffh(lm(8Ja@Qj9fh1@>F1dD(?9Ki;1;Rk4Zm%#niVy
znR57YHARe1$ao}o`5RMv<4C_KeEpwIOO$D*X8O^$1u-3dw0CMOnG&+E)bW&^)v$8^
znLF-IE(IAna&=%k!$*%@m(5?Vl&_IVHB9a;ufP+Z9kl63j&h^q4uzzlX>w&jgF6EV
zQrbSeWR3{w@K9W|&G5R>polN?Yb6TWN$*aTsPHEI&x%La+)k3)!1=5c#wJsi9y4Tw
z#)wW*F!gQ;KY!1z+h6VOR9{7BIoO1qatC8C1)ZM7x+o>9ae2aU%*CxvlaulcHt%c6
zYZlI%N|}6;6s4KaY`CbW;1v0-jl%b{r#(z|<Z0nF6~EgERqzv16!IX7pRohu<G>R+
zB~ZR%LykhC_WtqkU^7siZ2li6I7K_YRH4M7HC~k|OsuY)EB0>4X{Ozno-Gt7FF25m
z>qHLE(s8Nl!+iHLh0}Dz>(8-{Y@Dw12z1N+AeoP$HE8H3$%u8>kNXyEWms%JQF-)x
zyNyaES06Pi>#eSc5l%gZt^17++pt;*SW(K~dG4sKTg#7V(7079=I3y-!M$r$<6<QC
zLn{1jhkV)nh<aHomb2x>guuwUDu!{;syxe4TRg^%ADoxEcdhX5-Ha%XTf`HQ;45T6
z3w(!mrb<C{=8RebVN$56M%>ZDEz^3yk=)J?-Vosbrh%tgmfwacE22)sCWw}`k1^MD
zHUNgRyB<a^jMdk}aHZvthlIZdL(g_MY#RM_#RF+vpF>{e#3<XuMWim~o)x2p)v6&d
zWz|_@Ew(e#un<aA5-bhdhl)^wGoDmo?B<RVisGZ;PKO*(p-NMNgm~$Da4bb6k-QE4
z?{LYY%GG*0HFTBswU@co$Ypvx&G{?yJESP(An;2~t)#S9Zqcgy<0q*^dvKq((H#b9
zax$Qbj-GFLV&5xSGyU&{Rn!jjGtF<p_2?qoVEn6DV8Kf(oNbq`IU5?C{v#XX^IKE+
z0p9aJ7MTB~^wj|lg7|=^=kl8imJMXUKmmL?O1`@#A@$wZ#(j51P1l#|Vd7T5CGS^j
z{YlA(a8TRw&u8R{p!bh1C@Aj>{VmBQcA4al=2!-i4_W>QZ|n9Yd=Podv|-FBy~yex
z>CN@=9%}iGvu7;&MCzWr?_gmT=?uuDCq*fp1YIL|@>|V7RmwED2+i)P<MfkamCbrK
z`Da$DKr8d2We=$;h-QIS;VLuM6_NueXw|x#UZch^h!g>>Jr!x>!9W|Gl(xM7RBVuJ
z7Y<qzzKQ;KyG|&=t}8wE3YwU5fzDbkYF9n+;t~5L)z%{(qnw+QShiBT8JBKk+NI7!
z-3(@uP;{vqX=;GG8=hcwLkg5M6dg!0`AiNH@wozYiK2lTZcl}&sX6k6fZ=vKjZk&4
zU$qwEP#;E;rp}MK;dynL+u4SoH;=-sQ{!(vySr+G+V^UF+34Re{t%pv%o`#T1^lMI
z5d+b?ruhIUsp;FR06y~tddXN%9_|E!e}ZVS8N3!O59qz36%nUfnoE;kEzkM5jw~~f
z($4mj(=e7Vlh+<`j#Fmn)zf{tVnh9pv7#$>J~-BtwOzQ7G|xn1y0&mRTF?IN$>MjJ
z;)vz5T1T2y`z~yjefGWNuJCM(i~4Z#jh+u<#Bx7ZxbpBTL2XxhMT2ZiGwqo7g$%)f
z>>95{1W~kOJXNI8UENLTj!3zrXi&sGaGqFn;&ZKu<smTd*S>7MSH&Q31AX|Y2&?;9
zn`va%7k@{O9*zfPl0I4Z;)2z-43@pub@4s8idH>LI&_A01;K_b)es8cz6UmyMW^wg
zqT$@OybqZZlaoG((in`S_|)RanfkH*jw3tmKfE&J4ef_b3~TQg#mQN)hZCw$UsIss
zXI2#l&*_^+o31MFgHt7%sImK2PAuZeP2g8!#6}_wpfv!g2=Eyh6{M4EZcT_Mosv8A
zxeGb;a^Puc`=!S6mfu_bZN18*x^j9fpi3i%C~D|&D8((?-0G>HYMM#UhBef0<H(Eg
zJb(r+qQQdD^w~=gMO!`(Wons5TfB@>1xRH=(b>7%bow7A-Uy77#z#c^l9zcE>L=u8
z5DPdJk`vnxp^)P&Sg+%B-9ktyrWL{RR1)1TEh#}P&z2h!1Z(=wO`5ylBtMO-(~ABM
zhgDsSXuMTtps%k-tt<r(8xx5ftJI@&!rFdQwCF|hW%7q3F!90Qi^dX?4^1uj((zfp
zFHsLBs5ms5?|KE;Ek&q%c;3m~bJe%nVjU;VDoiLD-D&%Mee7Ip#QM_<udRxF`M+@C
z{=Ul*&4(w|Z#6X3fg$nx)h9s6QO2wMLr$E?kAMGOl>h3FsdvKaLx&?zghoR1|DhcC
zds_eFCsy)TR<h4^cyMJc(Yos!uSfZK^Q={H-lFp8;?;{Qq$x~oJX-E1h=$=EfgS5U
z&G5inLo0R)^ZlH0a+8U2(l2QR){R4J0n`nqW?wZ*1)ouk+YIdKZB0sG_37JI?P-89
z^;k!;N$oQFBTF}-dO+H2OZdTU{@Pfb`kSA<FHm1KMWk?UuRX>OLxiSjK!hRIu1NHj
z_OO0>-MO}v)HM@Vg}H4bQswI5or~M2dS7(jyk=w`ajzhu!J0WRkGteOpB7sxYaePc
zzb`PSUyohYBI(#Ehrjsp`{=2Q2pI-Dm&6_=7Id9M!y_M&C9X#z>?!GSBpLQWnZn7t
z(J2FB<#AL((*h?0pGDsC4WNyN(2~eR*;Dz_vlJM{h&I)PJf)jSi@LdU_8P3Su2){8
zf&G&fmc>!P;-$hfa#w#HoMordu74v&>dgFeWSUO4Z}Gr4m^%LB_?f4!`1~~s>y8)`
zE!2CZ^X8~tgcXa#Il=bVCd2iHVBnI;KLeY9fq7YV&&q0YHavi<pW=3}#50*iT!T)@
z7p0`J_4(T~Dw5$aex;7VX|LUQA{Uf(Xm%Uq82Dv{8*8BE5&N!~({7IJDT1TEs%oRd
z=OXv}9Ut~7CbH)6geTkP7G|s~*VYXt8{P3?6$yULb2v7i7C%hjFJ95^dsE*>VPB$w
zvNmRmlCI%gg7Lv6WNmj8P0u!%GA-`K3=}5h-AXu-2ELRe^g<KVXPMmn*HhIz_!q?)
zQW3!l)>}?oY4+#3Qz%{eEX9;!6HfVHp7?l-b7T*sui|n^Yj2BFwW$T>){^+^;H%ja
z>rMJROfe>JxEkKo@p~(BCU;NdI%z7*c{&?4%6rkj(Qj3}Z?BDQ&oL7b6QcCx&LBkM
z%=!9PwRvab<j(8kh@Mv4il!#G2gL~~w{ZWM7`e~{Z6rEMQi)LWY(BB^VJWj%fYhJ^
zXMs+jp#EHbrvR@FrOPBnO>72_0i24MRfs_n>T{}{l9`uE@I=;`WF<~C*F|)`w;aoL
z#nVcwlD==e*u#0Uy#%nvYUf!XGXoU%i{4h#X+NWycOQ~2aTO*^)ypw4Ym>ClbhEI8
zE=Y3*TdWu*^>92AR5G{XH}eme*tBys()wXk#?PCBOad_{E{)>X-Ez+SF3Jh}3^Y{2
zaXOQjRTTbXx}tEG{|8w?vg%uOemKGYWc`F);STwkWIK_sTuGT;-8poy)zvM}#h54T
zPc(I>K|%keX-ew+jJ*4@{5!leXm3qn0Uv7R-hki$0W;o-P*d%jn!c<MtphoR)6d8c
z-o@WFt4x0A(!Te4Fz$vqg>wOCTAOu-&cf}&N)ok6xmZIMDf0KNCn}#_*Fvs{2XdR*
zW-^d2QY{jjc5P*d6QAr*_Tktv!6swFy^($_Z8V)nFhlvRz*z3Q6z3Bk5?4qbsffqe
zSR=M7ifI+klT~Kd%c}tYtD|8+!%->SC^MwsQLT-6<NiaP3V~u?_>$g0TqY;VfuH%7
zQMAF!99K&5m7<1=Z7F`IFe*mO2?g8_x6WvQ!~(x{AkNY?<7gBc=b#8Cbx8g@d7ge!
zEGe5#^(mhIK#2VuZ&rK5KExdLe9M=+T%=Zn3XAf`*p+R9g@bZa2@8;L!*nZf19U9o
zh6NZeUd&LB{Io*sMw%DNBfA&je3|5s-hd#K`FZP%TH3{FiH2NpUM)|hnYXw#Z}&!4
zG}V)iwjvwezGRIhNK?=JwSq@i4*AQ3*?25>qf*xxPi9AHtk_6-R#_{O9350}M{pB{
zmVUqyna_MmnK6ag<C-T)w=}2vWP1bsb!i@GXnc-leU#l<tPby3$1)-%oZbaaQK?hv
zrx_ifFHBe`kDtERXR;~k?=#R!ITj$!T#Z+|JNrtcw#vk{&Ub@tM*1h#B~+4UnYp03
zR^8mJS5G!Jg7o0pJ6tWBc$$9$N@OCQdW@W@<2{dMQ1st36ka{O+)2wiARQ;I4Ckk>
zo_h#c4Tm;e9FE%;I#=eT8{-)0qr0k?gzG0QOZ-u~{~&)+^6?thh3z)$jB@3gU3}v>
z&Rp6a3x(ZnsnEQ8zM$DgdpeCQ_LdlqBHX{F!cFahw69+PWg5eXL_+{ZB-3_Tz2LGR
z45vX-`^cb*M`n7oMN>M@MCNN3k}pl_Ufo4%ReU1*+em3EeoI*hTR7Xv+2gY=J?EW2
z|Ae(Yp+{z}3{z}d4;d2k-gjjk=c9c|A(3oxC(%$JF{oNtTzo3NR}HRkIbdAw=~@rY
zKr8Q<MY@RsQ&AA|&fs?XUG4Bqdd`9+6IyNOV3jCD;{lYp9LW__>ew59D^F1Z!21Cn
zeST5)Ep)m%1*nfjtr`Uv^^9`lZt^RA#o1Um!nZ`XP6KoP5`{^87n8!?C1B!XPM?tg
zM*HfV=tyTT0@(j^7g=~uDBh*vVh^7-nlUemvIv}f{@+%`|CJ;0ZS$nKJ|z|Ti4XNZ
zGcW%6PPYHXfAanDT}tokJ+l1y+pgiK{4=6#=&xUAL?81&#*(Lz;GJ`)bhV7G;WQi~
F{}-v9i)sJ>

literal 0
HcmV?d00001

diff --git a/docs/source/index.rst b/docs/source/index.rst
index d985dfff..310274a7 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -111,6 +111,12 @@ If you use the code, datasets, or models in a publication, please cite the appro
 
    replication/replication.rst
 
+.. toctree::
+   :maxdepth: 2
+   :caption: Enabling OpenVINO™
+
+   openvino/openvino.rst
+
 .. toctree::
    :maxdepth: 2
    :caption: Benchmarks
diff --git a/docs/source/openvino/openvino.rst b/docs/source/openvino/openvino.rst
new file mode 100644
index 00000000..c60322a5
--- /dev/null
+++ b/docs/source/openvino/openvino.rst
@@ -0,0 +1,72 @@
+Enabling OpenVINO™
+~~~~~~~~~~~~~~~~~~
+
+This tutorial introduces how to enable OpenVINO™ for GQCNN deployment on Intel® devices.
+
+Intel® Distribution of OpenVINO™ (Open Visual Inference & Neural network Optimization) toolkit, based on convolutional neural networks (CNNs), extends computer visoin workloads across Intel® hardware (including accelerators) and maximizes performance. The toolkit enables deep learning inference at the edge computation, and supports heterogeneous execution across various compution vision devices -- CPU, GPU, Intel® Movidius™ NCS2, and FPGA -- using a **common** API.
+
+.. image:: ../images/gqcnn_openvino.png
+   :width: 440 px
+   :height: 250 px
+   :align: center
+
+Install OpenVINO™ Toolkit
+=========================
+The toolkit is available from open source project `Intel® OpenVINO™ Toolkit`_.
+
+.. note:: GQCNN uses two layers, RandomUniform and Floor, which are not supported by the 2019_R3 release of OpenVINO™. `PR #338 <https://github.com/opencv/dldt/pull/338>`_ adds the support for these two layers.
+
+You may get start with `Build Inference Engine`_. The ``Introduction`` section lists supported device types. The ``Build on Linux* Systems`` section tells how to build and install the toolkit. Here're the CMake options for reference. Need adaption to your specific environment. ::
+
+  $ cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local -DGEMM=OPENBLAS -DBLAS_INCLUDE_DIRS=/usr/include/x86_64-linux-gnu -DBLAS_LIBRARIES=/usr/lib/x86_64-linux-gnu/openblas/libblas.so -DENABLE_MKL_DNN=ON -DENABLE_CLDNN=ON -DENABLE_PYTHON=ON -DPYTHON_EXECUTABLE=`which python3.6` -DPYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython3.6m.so -DPYTHON_INCLUDE_DIR=/usr/include/python3.6 ..
+
+Then install the ``Model Optimizer``. ::
+
+  $ cd model_optimizer
+  $ sudo pip3 install -r requirements*.txt
+
+And setup environment for the toolkit. ::
+
+  $ export InferenceEngine_DIR=<path to dldt>/inference-engine/build
+  $ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path to dldt>/inference-engine/bin/intel64/Release/lib
+  $ export PYTHONPATH=<path to dldt>/inference-engine/bin/intel64/Release/lib/python_api/python3.6:<path to dldt>/model-optimizer:$PYTHONPATH
+
+Freeze a GQCNN Model
+====================
+A frozen graph file is expected by the Model Optimzer of OpenVINO™. This is done in ``gqcnn.model.tf.GQCNNTF.initialize_network()`` right after the graph is built. Seek into the code pieces below. ::
+
+  # Freeze graph. Make it 'True' to freeze this graph
+  if False:
+
+Switch the ``if`` condition to ``True``, run an example policy with a specific GQCNN model (refer to ``scripts/policies/run_all_dex-net_<model_name>_examples.sh``), the frozen graph will be created into the file named ``inference_graph_frozen.pb``
+
+Convert a GQCNN Model
+=====================
+``mo_tf.py`` is the Model Optimizer script for converting a Tensorflow model. ::
+
+  $ sudo python3 <path to dldt>/model-optimizer/mo_tf.py --input_model inference_graph_frozen.pb --data_type FP16 --output_dir <path to gqcnn>/models/OpenVINO/<model_name>/FP16
+  $ sudo python3 <path to dldt>/model-optimizer/mo_tf.py --input_model inference_graph_frozen.pb --data_type FP32 --output_dir <path to gqcnn>/models/OpenVINO/<model_name>/FP32
+
+Parameters passed to the conversion script:
+  #. ``input_model`` the frozen tensorflow model to be converted.
+  #. ``output_dir`` the directory of the converted model.
+  #. ``data_type`` data type of the converted model.
+
+For more detail instructions on model conversion, refer to the `OpenVINO™ Docs`_.
+
+.. note:: ``gqcnn.model.openvino.GQCNNOpenVINO.load_openvino()`` expect to load an OpenVINO model from ``models/OpenVINO/<model_name>/FP16``, where ``model_name`` comes from the original GQCNN model, e.g. ``GQCNN-4.0-SUCTION``, ``GQ-Suction``, etc.
+
+Evaluate the OpenVINO™ Model with Example Policies
+==================================================
+Now the GQCNN model has been successfully converted into OpenVINO™ model. You may evaluate the GQCNN OpenVINO™ model with the example policy. Seek ``cfg/examples/replication/dex-net_<model_name>.yaml`` for the below configure: ::
+
+  # openvino: OFF|CPU|GPU|MYRIAD
+  openvino: OFF
+
+Toggle ``openvino`` among CPU, GPU, or MYRIAD. This configure specifies the target device type for the GQCNN inference to execute (supported device types listed in the ``introduction section`` of `Build Inference Engine`_). Then run the example policy in the same way as given in ``scripts/policies/run_all_dex-net_<model_name>_examples.sh``, e.g. ::
+
+  $ python3 examples/policy.py GQCNN-4.0-SUCTION --depth_image data/examples/clutter/phoxi/dex-net_4.0/depth_0.npy --segmask data/examples/clutter/phoxi/dex-net_4.0/segmask_0.png --config_filename cfg/examples/replication/dex-net_4.0_suction.yaml --camera_intr data/calib/phoxi/phoxi.intr
+
+.. _Intel® OpenVINO™ Toolkit: https://github.com/opencv/dldt
+.. _Build Inference Engine: https://github.com/opencv/dldt/blob/2019/inference-engine/README.md
+.. _OpenVINO™ Docs: https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html

From fa923de5ad968214882b3cf041abdecd126f2ecf Mon Sep 17 00:00:00 2001
From: Sharron LIU <sharron.xr.liu@gmail.com>
Date: Fri, 6 Dec 2019 17:23:05 +0800
Subject: [PATCH 3/4] Enabling OpenVINO: freeze GQCNN graph
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

GQCNNTF.initialize_network() is updated to freeze the GQCNN graph,
right after the initialization of the network.

This code block is inactive unless explicitly switching the `if`
condition into `True` when freezing is needed.

A frozen graph is expected by the Model Optimizer of OpenVINO™.

Signed-off-by: Sharron LIU <sharron.xr.liu@gmail.com>
---
 gqcnn/model/tf/network_tf.py | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/gqcnn/model/tf/network_tf.py b/gqcnn/model/tf/network_tf.py
index 8b90d135..d2a86551 100644
--- a/gqcnn/model/tf/network_tf.py
+++ b/gqcnn/model/tf/network_tf.py
@@ -552,6 +552,19 @@ def initialize_network(self,
             if add_sigmoid:
                 self.add_sigmoid_to_output()
 
+            # Freeze graph. Make it 'True' to freeze this graph
+            if False:
+                from tensorflow.python.framework import graph_io, graph_util
+                self.open_session()
+                frozen = graph_util.convert_variables_to_constants(
+                    self._sess, self._sess.graph_def, ["softmax/Softmax"])
+                graph_io.write_graph(frozen,
+                                     ".",
+                                     "inference_graph_frozen.pb",
+                                     as_text=False)
+                self.close_session()
+                self._logger.info("Wrote frozen graph")
+
         # Create feed tensors for prediction.
         self._input_im_arr = np.zeros((self._batch_size, self._im_height,
                                        self._im_width, self._num_channels))

From a3eccff2ee75eaaf090e2af7c00fdf8760bebc8e Mon Sep 17 00:00:00 2001
From: Sharron LIU <sharron.xr.liu@gmail.com>
Date: Fri, 6 Dec 2019 17:22:19 +0800
Subject: [PATCH 4/4] Enabling OpenVINO: added GQCNNOpenVINO
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

GQCNNOpenVINO, derived from GQCNNTF, implements GQCNN inference with
OpenVINO™ backend.

GQCnnQualityFunction.__init__() is also updated, to load the
GQCNNOpenVINO model.

Signed-off-by: Sharron LIU <sharron.xr.liu@gmail.com>
---
 gqcnn/grasping/grasp_quality_function.py |   7 +-
 gqcnn/model/__init__.py                  |   4 +
 gqcnn/model/openvino/__init__.py         |  14 ++
 gqcnn/model/openvino/network_openvino.py | 222 +++++++++++++++++++++++
 4 files changed, 246 insertions(+), 1 deletion(-)
 create mode 100644 gqcnn/model/openvino/__init__.py
 create mode 100644 gqcnn/model/openvino/network_openvino.py

diff --git a/gqcnn/grasping/grasp_quality_function.py b/gqcnn/grasping/grasp_quality_function.py
index 5d5637e2..61ac237b 100644
--- a/gqcnn/grasping/grasp_quality_function.py
+++ b/gqcnn/grasping/grasp_quality_function.py
@@ -928,7 +928,12 @@ def __init__(self, config):
         self._crop_width = config["crop_width"]
 
         # Init GQ-CNN
-        self._gqcnn = get_gqcnn_model().load(self._gqcnn_model_dir)
+
+        self._gqcnn = get_gqcnn_model().load(self._gqcnn_model_dir) \
+            if ("openvino" not in self._config or
+                self._config["openvino"] == "OFF") \
+            else get_gqcnn_model("openvino").load(
+                    self._gqcnn_model_dir, self._config["openvino"])
 
         # Open Tensorflow session for gqcnn.
         self._gqcnn.open_session()
diff --git a/gqcnn/model/__init__.py b/gqcnn/model/__init__.py
index d120daa1..bd21beaa 100644
--- a/gqcnn/model/__init__.py
+++ b/gqcnn/model/__init__.py
@@ -60,6 +60,10 @@ def get_gqcnn_model(backend="tf", verbose=True):
     if backend == "tf":
         logger.info("Initializing GQ-CNN with Tensorflow as backend...")
         return GQCNNTF
+    elif backend == "openvino":
+        from .openvino import GQCNNOpenVINO
+        logger.info("Initializing GQ-CNN with OpenVINO as backend...")
+        return GQCNNOpenVINO
     else:
         raise ValueError("Invalid backend: {}".format(backend))
 
diff --git a/gqcnn/model/openvino/__init__.py b/gqcnn/model/openvino/__init__.py
new file mode 100644
index 00000000..33eeeaa8
--- /dev/null
+++ b/gqcnn/model/openvino/__init__.py
@@ -0,0 +1,14 @@
+# -*- coding: utf-8 -*-
+"""
+Copyright (c) 2019 Intel Corporation. All Rights Reserved.
+
+GQ-CNN inference with OpenVINO.
+
+Author
+------
+Sharron LIU
+"""
+
+from .network_openvino import GQCNNOpenVINO
+
+__all__ = ["GQCNNOpenVINO"]
diff --git a/gqcnn/model/openvino/network_openvino.py b/gqcnn/model/openvino/network_openvino.py
new file mode 100644
index 00000000..f746bdb8
--- /dev/null
+++ b/gqcnn/model/openvino/network_openvino.py
@@ -0,0 +1,222 @@
+# -*- coding: utf-8 -*-
+"""
+Copyright (c) 2019 Intel Corporation. All Rights Reserved.
+
+GQ-CNN inference with OpenVINO.
+
+Author
+------
+Sharron LIU
+"""
+from collections import OrderedDict
+import json
+import math
+import os
+import time
+import numpy as np
+
+from autolab_core import Logger
+from ...utils import (InputDepthMode, GQCNNFilenames)
+from ..tf import GQCNNTF
+from openvino.inference_engine import IENetwork, IECore
+
+
+class GQCNNOpenVINO(GQCNNTF):
+    """GQ-CNN network implemented in OpenVINO."""
+
+    BatchSize = 64
+
+    def __init__(self, gqcnn_config, verbose=True, log_file=None):
+        """
+        Parameters
+        ----------
+        gqcnn_config : dict
+            Python dictionary of model configuration parameters.
+        verbose : bool
+            Whether or not to log model output to `stdout`.
+        log_file : str
+            If provided, model output will also be logged to this file.
+        """
+        self._sess = None
+        # Set up logger.
+        self._logger = Logger.get_logger(self.__class__.__name__,
+                                         log_file=log_file,
+                                         silence=(not verbose),
+                                         global_log_file=verbose)
+        self._parse_config(gqcnn_config)
+
+    @staticmethod
+    def load(model_dir, device, verbose=True, log_file=None):
+        """Instantiate a trained GQ-CNN for fine-tuning or inference.
+
+        Parameters
+        ----------
+        model_dir : str
+            Path to trained GQ-CNN model.
+        device : str
+            Device type for inference to execute CPU|GPU|MYRIAD
+        verbose : bool
+            Whether or not to log model output to `stdout`.
+        log_file : str
+            If provided, model output will also be logged to this file.
+
+        Returns
+        -------
+        :obj:`GQCNNOpenVINO`
+            Initialized GQ-CNN.
+        """
+        # Load GQCNN config
+        config_file = os.path.join(model_dir, GQCNNFilenames.SAVED_CFG)
+        with open(config_file) as data_file:
+            train_config = json.load(data_file, object_pairs_hook=OrderedDict)
+        # Support for legacy configs.
+        try:
+            gqcnn_config = train_config["gqcnn"]
+        except KeyError:
+            gqcnn_config = train_config["gqcnn_config"]
+            gqcnn_config["debug"] = 0
+            gqcnn_config["seed"] = 0
+            # Legacy networks had no angular support.
+            gqcnn_config["num_angular_bins"] = 0
+            # Legacy networks only supported depth integration through pose
+            # stream.
+            gqcnn_config["input_depth_mode"] = InputDepthMode.POSE_STREAM
+
+        # Initialize OpenVINO network
+        gqcnn = GQCNNOpenVINO(gqcnn_config, verbose=verbose, log_file=log_file)
+        if (device == "MYRIAD"):  # MYRIAD batch size force to 1
+            gqcnn.set_batch_size(1)
+        else:
+            gqcnn.set_batch_size(64)
+
+        # Initialize input tensors for prediction
+        gqcnn._input_im_arr = np.zeros((gqcnn._batch_size, gqcnn._im_height,
+                                        gqcnn._im_width, gqcnn._num_channels))
+        gqcnn._input_pose_arr = np.zeros((gqcnn._batch_size, gqcnn._pose_dim))
+
+        # Initialize mean tensor and standard tensor
+        gqcnn.init_mean_and_std(model_dir)
+
+        # Load OpenVINO network on specified device
+        gqcnn.load_openvino(model_dir, device)
+
+        return gqcnn
+
+    def open_session(self):
+        pass
+
+    def close_session(self):
+        pass
+
+    def load_openvino(self, model_dir, device):
+        self._ie = IECore()
+        # load OpenVINO executable network to device
+        model_path = os.path.split(model_dir)
+        model_xml = os.path.join(model_path[0], "OpenVINO", model_path[1])
+        model_xml = os.path.join(model_xml, "FP16",
+                                 "inference_graph_frozen.xml")
+        model_bin = os.path.splitext(model_xml)[0] + ".bin"
+        self._vino_net = IENetwork(model_xml, model_bin)
+        self._vino_net.batch_size = self._batch_size
+        assert len(self._vino_net.inputs.keys()) == 2, "GQCNN two input nodes"
+        assert len(self._vino_net.outputs) == 1, "GQCNN one output node"
+        vino_inputs = iter(self._vino_net.inputs)
+        self._input_im = next(vino_inputs)
+        self._input_pose = next(vino_inputs)
+        self._output_blob = next(iter(self._vino_net.outputs))
+        self._vino_exec_net = self._ie.load_network(network=self._vino_net,
+                                                    device_name=device)
+
+    def unload_openvino(self):
+        del self._vino_exec_net
+        del self._vino_net
+        del self._ie
+
+    def predict_openvino(self, image_arr, pose_arr, verbose=False):
+        """ Predict a set of images in batches
+        Parameters
+        ----------
+        image_arr : :obj:`tensorflow Tensor`
+            4D Tensor of images to be predicted
+        pose_arr : :obj:`tensorflow Tensor`
+            4D Tensor of poses to be predicted
+        """
+
+        # Get prediction start time.
+        start_time = time.time()
+
+        if verbose:
+            self._logger.info("Predicting...")
+
+        # Setup for prediction.
+        num_batches = math.ceil(image_arr.shape[0] / self._batch_size)
+        num_images = image_arr.shape[0]
+        num_poses = pose_arr.shape[0]
+
+        output_arr = np.zeros(
+            [num_images] +
+            list(self._vino_net.outputs[self._output_blob].shape[1:]))
+        if num_images != num_poses:
+            raise ValueError("Must provide same number of images as poses!")
+
+        # Predict in batches.
+        i = 0
+        batch_idx = 0
+        while i < num_images:
+            if verbose:
+                self._logger.info("Predicting batch {} of {}...{}".format(
+                    batch_idx, num_batches, num_images))
+            batch_idx += 1
+            dim = min(self._batch_size, num_images - i)
+            cur_ind = i
+            end_ind = cur_ind + dim
+
+            if self._input_depth_mode == InputDepthMode.POSE_STREAM:
+                self._input_im_arr[:dim, ...] = (
+                    image_arr[cur_ind:end_ind, ...] -
+                    self._im_mean) / self._im_std
+                self._input_pose_arr[:dim, :] = (
+                    pose_arr[cur_ind:end_ind, :] -
+                    self._pose_mean) / self._pose_std
+            elif self._input_depth_mode == InputDepthMode.SUB:
+                self._input_im_arr[:dim, ...] = image_arr[cur_ind:end_ind, ...]
+                self._input_pose_arr[:dim, :] = pose_arr[cur_ind:end_ind, :]
+            elif self._input_depth_mode == InputDepthMode.IM_ONLY:
+                self._input_im_arr[:dim, ...] = (
+                    image_arr[cur_ind:end_ind, ...] -
+                    self._im_mean) / self._im_std
+
+            n, c, h, w = self._vino_net.inputs[self._input_im].shape
+            input_im_arr = self._input_im_arr.reshape((n, c, h, w))
+            res = self._vino_exec_net.infer(
+                inputs={
+                    self._input_im: input_im_arr,
+                    self._input_pose: self._input_pose_arr
+                })
+
+            # Allocate output tensor.
+            output_arr[cur_ind:end_ind, :] = res[self._output_blob][:dim, :]
+            i = end_ind
+
+        # Get total prediction time.
+        pred_time = time.time() - start_time
+        if verbose:
+            self._logger.info("Prediction took {} seconds.".format(pred_time))
+
+        return output_arr
+
+    def predict(self, image_arr, pose_arr, verbose=False):
+        """Predict the probability of grasp success given a depth image and
+        gripper pose.
+
+        Parameters
+        ----------
+        image_arr : :obj:`numpy ndarray`
+            4D tensor of depth images.
+        pose_arr : :obj:`numpy ndarray`
+            Tensor of gripper poses.
+        verbose : bool
+            Whether or not to log progress to stdout, useful to turn off during
+            training.
+        """
+        return self.predict_openvino(image_arr, pose_arr, verbose=verbose)