From b84f6b5536c008c8a3c9615d7d52d935921fe43c Mon Sep 17 00:00:00 2001 From: Daniel Wiesmann Date: Tue, 28 Feb 2023 15:57:05 +0000 Subject: [PATCH 1/4] feat(post): add alquimodelia post --- _config.yml | 8 ++++++ _posts/2023-02-28-our-models.md | 49 ++++++++++++++++++++++++++++++++ assets/images/joao.jpeg | Bin 0 -> 24338 bytes assets/images/our-models.png | Bin 0 -> 21756 bytes 4 files changed, 57 insertions(+) create mode 100644 _posts/2023-02-28-our-models.md create mode 100644 assets/images/joao.jpeg create mode 100644 assets/images/our-models.png diff --git a/_config.yml b/_config.yml index a226c26..b3a4300 100644 --- a/_config.yml +++ b/_config.yml @@ -36,6 +36,14 @@ authors: web: https://github.com/vitornvpaixao twitter: https://github.com/vitornvpaixao description: "Frontend programmer and forever learner. Has been in charge of maintaining and extending our Vue.Js app." + joao: + name: João Santos + display_name: João + avatar: 'assets/images/joao.jpeg' + web: https://github.com/JotaFan + twitter: https://github.com/JotaFan + description: "Versatile programmer, deep learning modeling wizzard, and geospatial analyist. Has contributed to the pixels platform and executed many projects for Tesselo." + # Plugins plugins: diff --git a/_posts/2023-02-28-our-models.md b/_posts/2023-02-28-our-models.md new file mode 100644 index 0000000..7dcf438 --- /dev/null +++ b/_posts/2023-02-28-our-models.md @@ -0,0 +1,49 @@ +--- +layout: post +title: "Model Alquemy" +author: joao +categories: [ AI, code ] +image: assets/images/our-models.png +description: "An introduction to Tesselo's AI modeling, explaining the model types we used for our mapping with EO data." +featured: false +hidden: false +--- +Tesselo's deep learning models are presented in this post. We have used them to +do large scale land cover modeling across the world. + +We have packaged our models into a repository that makes it easy +to use Tesselo's most common models. You can find the model references +in our [Alquimodelia](https://github.com/tesselo/alquimodelia) repository. + +Depending on the context and the goal of the modeling, we have used a series of +different models. They range from pixel based classifiers to time-series based +U-Net type architectures. + +## Use all bands + +For our modeling, we moslty used all available bands of the multispectral satellite +images. For Sentinel-2 we used the 10 bands that have 10m or 20m resolution. Similarly, +for Landsat we used the available bands. + +In our pre-processing pipeline we simply resampled all bands into the target resolution. +Usually this meant to upsample the lower resolution bands to the resolution of the +band with the highest resolution. That is 10m for Sentinel-2 images for instance. + +## Classifiers + +Here we are giving a quick overview of the different model types and their use cases. +Detailed posts about some of the models will follow separately as well. + +### Pixel based time series classifier + +This classifier is quite small but very powerful for small training datasets. It is +non-sequential and based on one-dimensional convolution. It has two branches that +are detecting patterns in time series at different levels. + +### Single scene image segmentation + +2D U-Net or ResNet based. + +### Time series of images + +3D U-Net based. diff --git a/assets/images/joao.jpeg b/assets/images/joao.jpeg new file mode 100644 index 0000000000000000000000000000000000000000..35b1b243db1908226bc5708bc529c1eb2bba393b GIT binary patch literal 24338 zcmeFZWmp`|7B1R@d(Z?6?#|$D!QDLphQZxK1Pc({NpN=y!6CRqaQEQB9YU}>-rYPs{Z7?dt81;QUQ^xu77w!zKLBh68F?811Ofqaus`5om6A%@%i00} ztKXGy74fa58Gj<`}6&uqpZBL8U>h*lY@tph-R;YYKe|G;pt*1^N731Q{%f439~_8sXzFu~vWQ4R$6mmM;g zJlx;-mK;_A+fe@eJ^W|v9^HURJ}dxI016TkG7=&RGBPqMDhe6~J|+e_ItCFg-V^-i z#N=epiAhN*XxQi}s92~;Ng4PUSvWYkxw*;d1w{D4!fagJ;721MR8&+9bPPgFOhPav zDJA&-^Y+jN;GjIV2_8fRz~O-4aX=4U0L%dpBCMw$JNxeq1P70Rh=h!SiiQposKEwc zk%EUqfJa0`K!8cVgVh5FIEc8EoDxWQ>ZZt4&iG)zm@E`($;x&Djj>}IE;AQ@RJ5nh z2#JVk>F604nYej)`S=9{rCvzO$jZqpXliNe=<4YkK+K^QmR8m_u5Rugo?frL13m-> z1%C_)jg5;>Nc{9UDLFeQH!r`Su&DTJRdr2mU427iM`u@ePj6rU!1%=E)bz~k-2Cd= z`o`wg_RpQ&lhd>Fi_5ED*S{bAdi3+3?Jvjv2fuJ&e!(Fiz#||(`UQgXgzfM+2#A!N zNVpQ}$fnMCRA4_8e94%s%63$0E{$UXGnX;6r!?HFv?q_Q{qgMoon!w0N1px5v48tD z4`9H9V4Vk#1Bd|!kdOR5UPbc+@4?n;C`cr+@0cd#n^+g}oMsOj6g=-o!tu_`FSl>c zi%rcN>rYp#)KcpAQm2ax8=Csoajt$(9I2ap7U&;Xzn+qnQL15I~W< zO(dP`o{RlHZw{*&9G~F}EUtRHaej}d&v(KdS^fb+xP4OOHn`a;Ln7BM(vq=mh`aDE z%@>mTTjwMq+evG2LwEE(KBZ@$UDbQXVj`!3BGEKT>H(N86&l2|>9mDIIkYj7-(nWp zSg+ymOD0U%`>nxAW3m?A)!)N-QiuH9dIyC**7-o`DB;eg5k)d-D>Gdz+T^O_O+)+M zfs!?x8}doEP)ME`o@474@_c^82V?IWZT_TVt%DH*S&H=Q{Pn_yYIfT3cCuMNiqzYV z>H{cVkG=#U|5k%jH%YkLk>=boPmz$gzy-mKWJyr?W`$yg);qKYNMi4*hFs_~uMs2d z*>#4bzP->OD;*2eSPm{DOa14m_88Ki0}^`)SST%NT!5`|7F;((=OblkD5P#1!FAq` z*J9{BT9{e$?RURl-V+8qwPBky{%@vIhS+#0>*@+#-PgzsHhjHvj%1u)-;%eo$VC>8 zNgipQ=swRv$Ii427=C?uRkjt*&?lCr^`(vuiA8p~ms~=o%faj}$QGxOH$0&CyXy6U zhP0p&BLzo$FHaq9QB=|P>@;*anU*rl!#=jEuJ44 z>*S7iR*QD1h#tP%i)*@Dhd)peJhXKu$kL}4ks!J%o-lAi6~41R<$R?8sgi51N?zs# zC&+`t=LmPS49PA$be6)WWQ|7qT+E~?BSPS71@h|a&p5=la+q|}C(FoWWG9kYv~2pV zY^}a_S8hg&-X>f-hdUeF2W7ew5h9{#c@myqT~&yDh?o2Z5hyR|EHpnOT(7^^O5wFv zpjSAv7sj%wdSOXuh9tsxI*jiN8q$icP9@a=m*Fch^e1V~(`0f>H;cOJkKjnuM70u( z5x`gMs0}_#M&>VlSDQZ^6okgM96v$v5{1X(jpPSx*#d)+nAaoT^I)Y8bz3Je{4nDj za-7zpT){6JZI8KO!OJo&r4;N0Jj;wGm44w9@U>*K_jV|d7>*qzNh z_X0=4i*UxseaJ=8{`vJGD#}>6Yroj4F>CV&kIdWeu1C5Q@4n)SFk6iqD_QC{M|lQu zGVpU(VC_XT+>_qdI4~SzBEreX>=@DOa^kjE1&XB!OboutnE7SXhLIzz6`CX!2{qs-H+cSPeL7Wqjtr1X& zJ1Z8SX2viSFagBM%t~)*6}XT|h(TgIJtHYRL@5*ceycBH7H# z$zumZ`AI|+l7Krk|xdNY%}PTR1Yo8E0?a-uvB+7=6xXL{b=T&W~y{weEK zPjc8B_*lc3`o68gMg7t)doz>^o{~$oCp6>NhO)>}-pi@PQ>hnB(m!1vfH`RJ_fFB? z2cS%sm4!MymFqoSHNA@y5)=1J+o=y4!bSa+GtO*qxXv@7L;CNBZW?-A`{rQTvdl1B zzgH!J4iMg|sSM+Kxy84E+}o`59Q&*>v%XG$6GI+gbLz9hoJbmmQj03dcd8&zKi_(IZlHt_k-ol2(OlrWqZj7hy8?1xXQi z$_LwDllMcmzh?U{%Q9bpn3KBx)qvC-3t!?p)B^OvWoB`B)HsmtV>n~8sT2%e= zgST@2>9A_y-gy~Gi3z^zazZqA<>cES$DXrvGjK#Lf84qa;km=VybieaQ)1kAuR^E_Ohy{tNq1K zut+Lwsx8K(H6iBb-74Qr*Zr-Bo=*7N2=s`H>>imbM^4Cm;}YM`gd>cU{ZhuO$Lk#BGD#>FA_XCpOV(G zLE`YZ7%_>_v`pLwQ}hd_dIlHkOWW-RdS66ER|kl5ePmkE==*Jft!5N5?hxV97SsdH zb*sW0_vn-4)VNW|bTy~dV7(YW{CKtlraT8??|4?+W}frxZ&5O>$tiVe4X!v%KckSG z%-B!~MIDfQvVoEYh%j7>!6X0O?pqD}wK!$JAsZS+syY$i-aU-lWUWzI13Q-OiRqm! zn$)C0ms`EJ}hp&@Q7$>lKIoO&>~C z8U?Pt(gog-ICuMI3FOpiy!M#s#%bg1hMQ<=Cy^>m=c=b(;V;ZXvv;|YXTQamf5ad? zEUlv|k_*PIE{Vp}N_FTqaProo3vG18`4jI1|c=&Aqa1)LGci7nxyX7$3Uj#mt!(LkIXnu|)Q)rTK| z@giD#sd+_nZ4~oO=P2(gYk-SPb*WgRtF*01px3+7?9+Q|&TBDGa^80(1`4DDJ=#Xs z#)Di6Cw;tuJIlFP2u$100|dkIN>AfXIy&Loz*t=@hJG5?S;8 zW!oG=jB)p(c;R-l>y?KO$DqeWR+p!MZm`I1PKT=Z&j+CDc}CF7^wv8Y|8-S{bz7{o z5!fWJAR&C4n0lAdv~E$AXH>D^*00X97dDm9t_^k;QWZTT=v7Lwg}4Gv_bt^6pKIsTo{&=+8G z2A^gf`*pM~%NpQKsndt&%6eMV<0Vp~Q+cjNNc&$tOzaZyQ1Fz}jt)g;hcc_7$?)g# z;76zn=%M7+_uAx5E8UpoCS2o>%c|=e%~7RwY!gVz)p?f5e~Qt{yB5!EBA~9qSW;Cr zNL-`v1*|W(+1@A6?8PE#z=P$kw9!*of4p5OvjtBmn5%J{pwnqy1#c_ZRZX=T%z8EN zHig1%871aCx5ZQXI=sEKD|pA8ALG&*p&oB|Y(v|ib%`h0tl(yA+dla88^w_`f zt@S+E-ceU7Md@K`*WtZ?MYz&*ULMYk>y4*>lV`QcFsC8tSwv>?gkp>|i72a~LV{FD zoxx%ua;^{-mV#@N+~GG`T+~$IkP7dTSRHu0?YoRU-7>9bVhWjMAuh{l#TO-xu2BkX zD3@LqB&yGMCtWt&N*RcfR5+)rDb4T$B5Qiw^d@x+kc$=RWEJ8?p3!!FHF6Uu;?oAv zL3?j_*gwnKlv}7$TDnV7PY6%!s0*i#yEr-*hJQjP&{4s%v2S)wwi>f(0GzV@>^sGYf68L1hDPt6^3vA@8cTmkmH+^vmr0e5FtN^_;= z;)~e&Q*`@Yq2JrbbkSzkE6hU!;)wO+4Rw^CIAT%&}KH?8fOUu|Zn31Mk(ICd=niD~pbFp{e>!{-4Z| z+Y%1Sssr{}F3#z)u%Fd}Fs)%v_6=C>3xzmx(L{bOLrlY8$4t&5aca3nB}?FJD^7kJ zq}E9&c3)gW6|*B=&YcX=l+WkV01;Pl2NT_QHY(Ko4}DM^BFr zqIks@(~TS9OTR8%H%QrA;2pUdHt>YplVC!^%2X?`XH}l`vr^1*8u@bmaGsp7x6YO> zz4mPVRblGJhE_H-+KtCJSBqaDlZA(4+~FWx*Y!sV$P<73X&c7i%-xF3s;SVV-(K>C zDqPo>9Zi$Cq=g4S!XU&pb2|BRjO0KAjrM1nP-`>_XTOXTf=?P1q!{V0M>o zz``kW)4A{Wn;Tu8h{)Ul{NpFSmN4(r5xR?{cK**%PwDe{y9qk?&DZ}?x6A#df|)uZyL#s%wB-ryW)4pNl> zpq+Hvh-c6Re)0sHov)4_WWB%Dgh)c%1H!fY9BSjbP5YEAYXsOT7tXVPth?seE{h)R z=-2U)d|WHyS$Qs|5V8R#OVD+6X2|>UmDd)0{uGGHGmcf#CZf5_y(9)J-Y)*uHY^fV zk1{aUL*i$I&;-FW80ZC^f@?cHeDqiB4FS_a_a{RQ@Xe`+^p^ z_iW$i?r)YnDqFMR_G0YKHhS`xM8sM|X1upcLOy|-pRdD_YpCK7yWV2l^DO?dRO7uI zryF(IIKL#jXqQhQA}kM6YP4XHBib%lRi|}M-KVFRY^pDe@hG>R@+&GRQ1$PVZ9Ofk zU$J$#>~?)RtRD76N#|#D@6&zPRBhIPB7)&;@dPg$Z}jw8u*l` zcezbYP5J;ZCGNe53c!QSp<#Dxg;h-5D0zF`R_$J_9p7}e!g_H`QF@O=!DKFa87Q}o}m?UlJIH}_UKAKZqcC^|L`fR4RJ^# z^jaM7e)n#VPX6+;a5JTVKtrFdT4>|wc4 zhLoX!qG9K#SKjT>II>?~oTY8WVI=ZlOeD}0!COvx-!U!FEi*+0v%maJL>HsQVfIETy6z6J1tMHR2|eaXFE9Oyj%SV^PfZZ6&?Jr4kmfisOx*R|oWPjjeST?#ga=Xq&&!*!!a zaZON!VpwoI*3)97mGSgghXZX#S1j6>KEcUK)!_WDQ9U zX~kehTI2_JzZgbK65Anx^dZzz_0sYUK)7M7b1B!_U9#OaTI-w^lc z+EpK9ZO%;CF{wFEWb{;8)}NuQ1%G38^cRbon6H|aRE5iCsC8Jt*~n0yl04l|=5M1w zJh~{NZ{RfW7r*&+$4vFCpROh`v>o3{Z@%neL8dEQwkfI*_Oo}s01ZV6WjjS3;V&oY z)IrjX-T$%>K*lxf#ZxDm*(Zblo*H(I$x{vK=Me$$d)nL8cns|n0gF?Fhr);Ay*5o4 zvm^NB24T|n2X+gu5Mt7qJd^YVkSlCr*9hv>P$}T&s~oaz`A{}U!%y_9%;R*a#>=~K zt7$ALTEu^&O|~{kRQt|eksEX*k1DPDQNxoaN1me6_MojV7x6~B&~>L%vKDUtdI0iV zBG_x(OUp>0@h^WZ^VQ``4(#!M{2ep?+$UT1tey0ZQl-&kO@l_UQEyqWlALm&V2 zhsTjx7izp;RMT=CQ?KohxvCNcLgx*<-kU+=^1MPMZSevk1* z1k*1^HYG`;)_Tb!%$D;Q%0}og8)`jztSi~$QQyB$Fqq$HQ;f#)%4f7Dfu-ST@8lh` z!vZ(xd_Cag{mv$@>9$}`sY#$4jC3tmTu(cDu8O6El|})q57B32mYj;04J*68$90*G z?e1M+{mtr*v@~hStbKqx7N0qnap}YQ^@*LU&Uf4JEdB?;Ccc(m%wYO!Bp3_bCsoJA zd+9mqGOvZ1s$XAzAMUDf10N!?=im%aPO&v6iat&|GJ5~-8*P-(Q!F2}`WC1e^@dqH z8)>Ta;1}_bQc=kWY^xl7k6+s@35o) ziEogZG^5M#Lf)h*LS?ChP0#h#?Av*AQ{jrdkKiCxp-I+z^}?gBrCq=4?gx#LT1C}? z=e-xD>_tk)q#r*u2xK#2jAL_fFp>$M`Y6=klWE|ko@ot+GA;7zS&rvK1!xi)K-%HS zsXU*q z^y?9QtLJs@4Yjx@C9<%KZEoEw4r0Z^7hY!X>T^Hr=I_hqEtjoV zL*?37CwA_6l1i<}box!>m^*H_Q_p@f*)rFU|JV>cNDmgyd4^_5-#b6eZ1TAs*F5SfxXkljAo01FV&DliXOgTLXH9FzH#6Nu3GS`o z5xe4!OD4g!l0KWRpyECo^)Ti>o2ulbAsttp@Hvv?z~uC;Oo&cQ9PyAkX8~tne=%i* z2ov90hI)TDLR`PI)>Xp{N3$d}!QVa?yDUP?i<6V(?khwA-e%rB01+0PB^pKi#hd&kQcQBL+B79! z!_88xA#6DUL3Gk4Ox4GtmJT1#K2qhCh&{|btYUat zxw$zBva>t5vYDDYnnBqhj`r+crcUgfY#i)>u$Y&VDa01)MqviEw001oI%;aAqOdj> zq0;41=1_K$f?8S2dAmS0y;Zaz-nI|{b1E@W3SlonFMB6@sGBK;m%W{XtDu(%)gR-6 zF#K4|PDSxY#LZTON)JYrkaBc^QgE|zvvIJ(C>tJNDwquya|=O@7czfafPE98`n#!~ zo}O%;Tx^ammh7AY0s`zDV0JK=6(+&z`r5(G)Qi=@mHHnGFQBdv7i%XsYexr)M~kLr zj_z(ERIu~^)i8S}W#xYv{$Iwx-u|(*f23XAq&;BD|GS8}YQ1)XvTHzH9o=0ZP-zdS zgB$ff>gJGt={mW)*!^+99KsH@gWAKyTwxLC{MV@-86^LbestK<+TQ7p1}w7wl7iuX z3H>h{|9EPchk`F0A?}YSk$)jV^=L`Z+!11JF8Bvpm4Hyo0NosRIQ1Xa{DQ%^GF|%*Vwmz{w9`wJ_xqfa#lavzl^2 zOj$Wixp}yF%pedxZr*?NRb8xM!^qU`Kil(Y#~fw{!ppwyTYnp|2=Xwq0WCJJ8OzR<3`XF@;C}b zs9a4wpypJ67ecH|9W0@+>j~DUe~DTDFG6f40EL>G@qk%5I4rnXxjDI@tY%P7Ggb)L zlo!g!4~2qxx&P63b+mBvG2q+|DhsLF#L%E<3DnqR?tTmVcaBE4lpa2 zU+XV!5+|6MgF}#mgNhy2e)h*p<}X*+|3@$X)cwch^^Z90ntg=-N&%u2|K0u@f&WI} zzY+Lv1pXU=|3={de-ZfiO$5{dHq-Hh-6A}!!F`dJl6s}4p(-Pj48SpkxH?IwsVV)#`Td`dm(!2*WdN99eYEvI z*Z*%zFwJ2%3ox!Q1x!jD;^g86!zM7y>*?n7h(E(HK5W8f4#TA|%;ExTAPmnwmYe;B z_aCw4AN)ul2k>3AG^Aj@;lnV6<$qwa|G*F{7kiitH%x}c+};6ZA5QNtZ2pMfK4N=2 zcUas0Y>ysdKpnI-VMiL+P725Z@_;g+22cQIfIDCf*a2<;E9~e1D{%!hVD*y!5BkJ^ z^wnTmW-u*l00Pr^0XPEofaxE7;IR*2GO+F6v30fJ0{9K`=J>Ka}|Aun7Id z9e<3)sBv5DUiNQlgIngp$CpA+`!L~_~6|*p4 z!HtF)#1%YxyV$VxW+(VbehdV~DNi1^MW>4H0~@%iQfJ#>peujePgH$xM^uW0Q#9y! zzQF?fxMVjxp@WxYv}9po(EnJY|3~1tMJxTx&PzL<*`o&_{bvL>>=l*3X&SZnjv|G8 zoN;fpCDTECO309go{w?|rm@P#54QqD_`9pHvD7fRpBfbTQMPO%^FIxPuZZ+9Wz5Jk ztS+gQ=9uCL#aYa@TWE%gLqoF;-f1bMzw+9zB2{_D-QLJ7yw)S5*j^q)+H1FE=!lDZ zAPEc~vmc+Q;&5h0N9rS~&pw}N6lFNAV%C0ZFhR>Ys?3S#;k{tfZg?uCm%BarMNc77 zxxTFa*k0#K()YsVN8a91hWn{W#c{%VOU$ zGj}ceyy66l7@cXMPUgd25qx`N(jG$F)WhrXthx=mRxth6p;0nDcFXB@TIMzPM;AH_ zl_%@)re;5)H?lSdt51c{NsRhZs7ktR`J)i0i~EM!W4kO!IYgB|QN!O5TeRYmzRs9o z>1YVQ;3Y9TjAZ@T$iF$TYeXIBzLq2dBdwXd9!B(Ueq{)WIgZ8zb9uF?j2NTN?9Zl%c3FI(TK z-3v@qvP{qB_nj-HPZ6s%SF6}{TDlk&eZzxGn3l(kyaT}r@q^Xv6XN|;7k>Sq58-(L z-rFq|mzJ6x9BHY|Ftw$OsYuoga&KGwR$rg4qGur)rqSg}wI!jdRX}c`oS07}weVRa?-sjNWyn9J!wC6SgOyFPdLUR-sk_x7oxx%v z;hg=xPi67LeN70NEibmt@Ymhy`Ciyi;s0x?2s{D`B0S=s!Sc8S1_$AB1&q2WA|919 zHa=_tOlA8q5`m<~7}#Z%nubg2%0f&8 zQ8lb%jU#v}S7V$Jz6+$Y%U+{2Zn~7?jy5yhF|+k9->)%f`POQ|W1&o5uv=uzF%4){ zue`UdJ^uWDJZOR=qQ@>aV&ZPX}Xeb3<*iM3(5ZzaN zJg-ZYjzV1D~ej0{lQLs(@YM`=)RqopTme)(tbcsvaQ$G$X(y@z%{F)E?z03 zG5NI_u3zvh)O&H@2oKWxNzch+?72>O_jvTU#^O^OKd-yep~m@`zB^u154e?|d|VkV z)wK!oIMhvg)ohE@9r;bUgn;nNm+2L;GP&Acx{#5NAsbq+zrCcD(>uCrBCB4n+lu1X z(etqLAKw&Vb8Km{-h36(B#1!{-FlsDq(V&>`Br_YH>eGpXph~^RXVS3K!w|y)U1Z;`h>-M!VVALAPUbOSVrd(`kUwdfaa3M5qt@oD_Tf?{LhAKt-sPXmRd?V? z%s0h-CWIEGm7@K_ejvs{b~v@M*R6_!%!nDCj`IXTp)5Kjwe8p2Z#@avL}AV(Z;B#K zA%#Mbr1$O?LU6V<Xi_x7dyxJnu~iP~V8;IVbfI{;<5r&L(6(Cs ztrm=OG}IEG8z7%r?!T2Ypv(Cb;Z=7%@4XEr`3?_R{v_^TZO2ydQVGMRO1mtD*kx|w z?|sp?#q!CQL#DmyDVpWhL=pWwpA6qEXS#=b?W1GklhNhNR8qR%32Lx=gq@rTPLm)r zOYO<8iQ7%{F?m@xi^nJ>C8*ftL8T6!k;;6^55~mj)kUYI^?$BcI2tp)HF@)v>9riO zeO+y`hIHf-lTOXmbX8<^$ne7Wix13S5d2-v$FpN{0X;o^HxfT}3DU@{BQxoPEbg%2 zj$rIzws}HDjOo+m6@~q`4o}VmzUWw8$}bg+$GuIPUlP-Ywy&ZtHV;M6>ryyuStH$>OZ4&F5g;pn}^3bpz^Mk0hJL? z&_Zfn3C5y869sZv@_AddV9eR9RxEcuOAimQ%DtC%>M2Xz&vv=i+B_E)(0Hdrn-hF= z9jHcWKFxuD^2kFtWIQ$He9x%nxz&yuPx5WC1zXGprA$LPyAQLA@E$DhasRVK4jCTd z&%E$)h1_3x4-uC5V0jN8EcuxDkf@KbHQL9_R{zR*aF039s%Eg@rj(BMcvQ(x_d7RP zn$Hx=`*GJkU*noDUsfCvXimD4AW$sC7(>VMso)@eXsm2iQ$Pw|l1hqUflZu0a)T+P zzNTBYP2D^&;3@65ANznH`c>V)1us6=^L>GrKRn1%)WdV2qi-+r5q(Lo#uJee;YyV| zExyoL+uQoBRg!`Yd)K(_k2;1ng?2&PI@>mSA)rz_8>5r*Q0W^@)aRw{7p1&j_QxrP zD)HmbYKE9fFfk*X(-}Xu4?Fb%wn)PgLi@5zm){gyL&t0F7UF#BejOR%>YsdNkrC0G z5^%q!srFK2TG{5&x_S!J)yppeKS#G$)0`F7Eu=Y51<8$mb0Q5I z7cMz=!(h=`X?EHtNADJ5p@lE_Hr{%`xH33$c3n;q!=EjU5!2XFt|)7lce*b;GwNC+ zIU{Z_?JL6kBus|#S%%@t?R4#?w^hyq`9OAeh5+__9L$sTFx0B zYo(-4A~!7@aW5;$Gqb7Q9~{JMB=?M^?pUkKRd4y)E~r($KOl~Ps}sH=7_+JY>3a4_ zdP$zlHbV_}g%e`aj@tcVUr#UNhU3+IOoA`PX%UUyWEBZotai-YlpDR`Z}kwUsgt0+ zhG{l*u#jX&*;mm@u5P-1lCy^zS=2=yQ68BsC1&On)9e$;vnQVIrqtc?YZTJS(jdia z*Cx)P1W&un6npPGyr)7>f-@Fq=blla_fwbJAO%8nP)Iy#`UubyYKBvP4i-_^ zFe-@~zZ(ZQSy(?8YQ6FP4xVg?BxxO=hYv5qNtg zyf?QnfJ}M6cfS9UjMzd9KVZK)7LK=0A*Q2*lUhQa%>zI~_du!r{$?_>`{<|g!D$6a zjaS`)#*JWG6iK2BVt0F;v4|}JHTP?o9{){>2$b2y-id07q^AbuBbQExcEDDryG=0XDJudzhbDe{?c;07t}-DwmOrj+VbfE zn76mr_~nBGKcf&4!qWP*ZO_c3{y;MrrC26saJe}~{JT4`K|1#=$kjx|`gosP(6z?v ziMn`CR|FG(S>|tpPgoybv7c?rC*Ww=@n*h`vFZI}7Md0Hxk?WgsQs`Yp+_4oxCI@H z61A-;ug+Y9S^}_H$LzaeYXlcBB?1RF)GKetnq9eET)Gw=ftYh4lHXcG?4hC8MjVT1 zsD*XQIpOH;HyW}KZ}q$9Z7&l-Q?A5y2b1iUy035dI_1K|zB#Cvs)oh#K%O)Yod240 zrl9%ZfCYWZcki-oB+A80c6vQp6U=MWEX!6y{eHEwMH70yqaNX?R{n{{#+YGGvsSv{ z-lM=-c%x|H%x4IG-GPNq4#{bHTAOJ}Wt^%u60Mf>ERzrYd8YZP#!bqJt zy_4PQwwYxCmiwk3U);;`##K9~YdajlCQV)?H-P5DAhsF}snC|+7aZx_lYY8@4^qUD z8alDnCx#SJbsvhzDB8jTCI(h*jbaRT^|H{$K$@ie{`OGwH<+b}?Kw5cybSd`Sy_Pw zTpPCaF>C4j(F)nOEQBs$Z~Z~6m&U%N%_~NCp-&fCCd#(?8}^d8Vck~NLUvKs`cX2r-cSd zrT?0A^Do2jH)pb2%`jy9(N7)qEP#wQ5A=b6i9dWl30H*r?ePeg?}mt4@|PZW-XF_x z1U2^D9_RWXva_FigD(}mcaG_V~9PBwyQWUN#Zo{^gp1Xc1v? z4BHUV^>jM5N}0?m9{JicE5vww(q`4rq;ZZQ{=Ch^gEfKAu4`|1AW04-R`fh~$b?rL=82tKUA;?-Sx_~K9 z&@DeL^s4YT`6bT)gRqVGj_R^IdzE=rFz?C%YS+)N(ra0W`Uag-`fOh;^CKNhb&xdp zC8vtE1nRbQAVxX`8kk=rF(eM5;c@Doe&r@=W%;|Zll~z5aa5F|#2@~jwMYxjc&^F6 zRjO~iw9A;G(7Z}xEiLmK#R+|44tf?LyK*jxCFecgcks&KpdcvU{&i$&yo&p^-xrC(DXrW<2Js&83; zM910y&q7$lweX-~aG0TgW!t-@tYz6yRj<8Oki|0t5>SlnOZ5op+i$qKpk&XYP9FXH z@8sFHvULV(b1Vz?3GD>QbDr6u>aVRYEeOpVe)Jn!tK`xsNCpk?i8-b{<=6rg6&8dQ znC(Ox&sB6cc<%BT9{~Lv-OWD6h4N4F;p_&#t!2BcuXWFf1x6K4YhX;o{ERdnSE#lj z{A7UgQ#Egk=KJJW3nQ7lKElQTG*P?P@9zXVYvWyR>c=qh)EMY|v1&@5fmBB{Wbz{B z+y#p6Gr#vV3)wmyGORmQKlN2cRE_u6JQ8#6?hN*^DBG>Bp7hreA$8ZG;K7uML6ko2 zVtqa|q#3TcInyeYZMyM>p<B}>0hojS;B!l@V$tOSEFlA80 zg3>l)$0#J9?TWcg4Zx`G0dW_^RfUl6hRqEH>#CiS##MyIYGFAL5)EnSZo~CE?xXaG zxY4Jduf*nzrR@cFjc#SuKKdNI57uCltJc#^UI^2|nJLUH=v1w!B6pXl!ByH0Q4m9S z6%UVI%o)%s{I4V?G#Veg`;g>;mA=C4=4&$MUEKLm?2GcVcK23}92)Lz{ zA^t&ppAStcgz{$4BQE%JK3EVz=Sc+1Kc|!4jwnHN3&K>!SBI!Ss=1 z{i1WOZ{R{i)lI{bK&x)dD;^G%&7$e<7fEsnFT9G+X)<&?3eWe2_H4gh#*ZIbF43@G z8hkeUAXpk&VU61Wj9o_?k6L?d+t)_f($jEu;&v0uNi(nv^ev^=m~oxeRB(zIpfY&t z#EVTP0C!qfW6v^rWmC>3SY2}cCCbZK6fUY$Ez3kia|~52=S`OOJ;pk4(zCr*#q;$R z2<)DE4eRVbRre^zH``LO@E=#hx|c@!HdRq3Q`c0lQ@LO3DYu>T6hmEvZIdoN;AmX# zHN-&Wc2bKs_PiXxD22N%&a8R{An{NE8!ut6B?%TIrW;tf?XITScvDERS9WJpg`}a=exHI@+7}HqM6$ zYMbdYLD}C^{ zvvJ3~wT&f*CXuhFTDN4(HBFH-ZR0(p;Z0%4pTU+})_uJY6S1S=!>(d-&qQaeYzauMn-RR>KIR#=kNCkr zKx+sW@^rC;>dc*I5KqG$HZ`Kfetael|M*b#??jr{nuIuZP@g={u`1(H5O`>yCtAC1-d-{6pRmG~A56h8pVs3oxsJ!yM&1(&YR z-fj^3P1thP?GPL#li4;AR`fT2GduG!iQGSxzUGtjYGe$_qG)>tFDbo4jTTEl2m~Pz z3E|;l=n}>{ngn~zF$-otE#so;@cJSdo%G^4VwLsVm(B3!1bgMjXRY5cLZ&tEVRHrH zkSo3wnMf=AmU`b}_c=9WML2SzDkl%32jEq@8=^j4-2-4!v7Q_wQgDw%edd+2D7hMX zjClNA{5ZzD97N2Z#u#Wc-jtQ9qOW!2hd3Z3fuXH|I%O?(abx@DiK@VzPz0~SEwc> zi(Qd&@Nb@x&x|1d!dJyRx+5PU1e(2lW-xlyLWKiY+E=J5#;~ltpH_Jt@2GuuTu4E? z#H|wkAVwYadC36C(lTv8xN;<5%B~F5s=zt-`>=1YEs8d|UhPXrzJyIDe-fing~k;j zXVNEA{bTCKPj5jw!)A?b_VMa1ZV%P$w^1T(3|U1M2H;crLHgNLm5%5b|G5IPOd_`A@;)0)1jVy09zMJgPUFh<>JRCRxKWV+!H>ZmbVh!*)q3bxKGk3v-NdR#X)#sMYOSlgFYT$-Z}Ei+@u5`73gb zLg?cToRiqXEk(#rJC1hhkC`lJ;sritu`jU$*E~3<7!siMm9Yol z2vHDut77}+qAz7%;XYK}F*v6ev0h$H*jU=_b?ZNG75}`N#df5dkc5A+9-keGy!L}A ztH(OiP-74+s)eg*X_b$wW$!&dWVNZOlc+<(97Muy+l~6uC?TZ{mY?ZxVCnh)_QngA zpE-d)`8g)*xUwC4>}2&{uPm@hojyEc3AFU8px#1(&l6Oj6E9j^eKm;|sgYVw(|E!U zH>OtYOW`7=qut9zi*4BFD5zSvT?zPcOCx4}h%)TAB$T?#rmQZc664?0!-}!jVI?SM z#yAmwD@EJGcxE}#OC!$Qha__SNpK}xd?3EI>S^cC=Pyx&H50jm;rI(pz8Qe_t>rj<)>R)))x|N6D{K+K4NPbE#~|=P1sycY z85U4;#bby7tO9^b=mn|8y#lfzTF`QGK!YUWHG!dscsLM77aQ~5I?VCO`JFLdu0A*G z{AnK@dDqi~>;_Mv{MvZ=;nI5kn)4 zhkE}25pZq6^6QBWQ4wfAy}2?j3hd5iJ$d72YYBU9eUo|kUmS`ANp)oX{`=y~l|&X$@_awv z4AtJ1lg_la6r zT<}Jka%qj1{yX9gpO??C*}e!F#qBrLTYu@mUucd}=8_(7uayBoT`WTQod+SsuwYu9 zOf%=Ru11I`32bH)Lsqin_|jh@<6m2CB|y zYGU=mpoEZ}X8QBv`{3?DYJqUEr7~uN`CNH{kKmb4xBgwFy+sLm0c4x+ksCx(fl;+% zwVkg31X#E%ZZa6ZP6nhs*L~>E+Utk_G8<~?!0eIZoW_H=5i{k+R|k$9!Xm>3AR?{= zB~iozViZWANdd-nqkL9XCXEu9E(r&zz(_khnLN+QN)U-3H^ajPm@96C<@2Ah%vGs49pqCQW-*oFw}R=B&AVct zvGK@DDWTI?$h3z@lzBYYki3lI0=e(Xd~1A4fRsz7iN7U7zMf98rQ4$<^z~nS z+{6ez5%+&{-}q7lXei>p-TF-Oc(PuO}EA_D?m zo~C(nA(5f;`;T7KpxD<^Fuyt1*UCMxME3|!sTpsX z-VRgc9Jok;QX2x7qnMy7RhJo*$kJGK%%o#M1zyZW#^Caaur(-4CNhENFOHfJ9^2ST_G z+@)v6F7>m7AcYVXB4%99hu?ff0vKa);_PRKPZTOWXAfQE&VL>8(JOw6z&i_DoTd|s zc>$nGVK|LI8{tDY3;B>Gh-G`#GNsQ-ozsa&-0hECS zj&crmV$LQrXkGM#Vc*x*`6qxz5M_iIL@Sx4Rb?ha0M@}s2N|Rbt;M=AVMZh|!HySb zl*SqH-MToTX!2)g*8X`K&Ve)?O>|{E1I)yQy`=R)@w3kgI5YxDV3nXka8=k0#grN0 z$NuU7gGIpX$Bws+;+YqL9|>+eu8r~93W*iB9uwoNuf9>eAtxe;sEF`su3gep<0??J zo5<^muXLn<;L4&j3gI|=fO3YGyW;Kcj)v&d`QX+BWD=AHLU(7zcrK9-vl$;_hmh?K zeg6Q*=Z!$X6C`DhM&J9-Gb@HzZ>&2BmmjdSfs732!-LPCt~90wwN`x}$=3-qJDElS z=v;n9s32IN%_bYjta0`LICCc+Ua}kzC1`8sKAHZ!FcKqsKosAab;Y4Y^L6zr6Fh*_43N%y zStR zo8*K=EL;Aq?pE&&l)xbMsdP~?9rX(VF_rUf%Fi7AAGXe zeV;y#3=BRk-|L4Y5Jf^!EN`Dz&mn`s28yC1*0uOJlePdro9g7E++y?x8Vb&YWQ7$~ z%H=JD9g!m?+P18@)u21rW%Vf8qWtif=X-+g`k@z_PBk8z;i zJsA425EWQW2HPWVap84sjh?KFi~0A$hYbYPp1ouD#h8O7M$f~tW;c^}szVe6QVmOa zBhZQg$N>cGb$2rz1f^%-1RNElbfx6#17Ruz{tHXDft`q81TSleR1&rv2`C^42ng7( zt<|Rj87hC6$1(|2^Gf=^gMPFDLWV^A%zoVH;5}yjU^hRn*ENqGHTA&4;3%WgR?VP z-~0V|aVS+gV6ok2`=N<|MJOsc!swa4B0ea{nGvY|)UnpMsUQHr!HmpcHyvIikd0dg zv=HBmi;e)SnC@90N!3D-0CwAzmL}W9Ayq5VuEDcNN1n_!~h@=0RaI40RaI3 z0s{d6000000TCepF+ovbaeVCs9{k6zqHu z=0yOgr$7oR6C}VS2%lB~hGCo387qmsSz^gw~lF`)4z;$TScl@meX{EsJM<)xP# zgM;&&{{TvSlAvP=BNNN@<{&5(diXL>xvfOn`es)2w62>1ha zUx|++Sd=cu3>W|;=%g?wxlTG%e4E}jSR*E1i5XrDhN^sU7D8^Q3c(I$fFryCDN7qeZ5d< z{pTI};-aI9-F$bP$18P+Eg1 z{{A}lrgXF=YsC%(VF`Q(3#7@1?SL4HB3rdR4Vz3K%l+~oQzGOfIiV;0eFf8PNBY=Q~Ike?d0wbZ|w$?{gbYiWnTc)GgB0-mlqV9=IOrUd?@ zII{7NG(5KMfouX(LIA0w%@rqBC=%a+68rv6CtLzhlQaklovmFyQ)b`h8uQ7k0fh<} zK?_VtT0rYRnwbWXMG7gfm;!|w6C;HT5uia<_Jjk31goAC0Oi!A!YNlu9=?O5Wvomo zU<8P=idnjDA>wJXMJqO~89SZ0nHtYw2U@93U@KVQF$Pj4Pm~*2) z*@Wo}EheRX1u$K4t+e2+sW)-py z!HqgJA1h2oL80d3A#Kznz#t`;O|Jr4iW0>FiUKBZb24o7kja9ENY9g8~5aS zrqAjDjLoGMKk=Nxmay@`R_S21Bg9BwHrf60K{|!GVk$ui01TeHr{zV+S+SDx2aWIkWx_0821m(+vI2BbZ2Fue$c zXt!e3$H;Lsi5v(~#~2_Aq42z@g1N!~*kJ(VObt!5+S|&>xwr_Pk(6Mg2u6m|l?&Tf zktVaXzO9KTLMs9)`&eWC@Or`#S{4IBHv9F7@#Gf=kVc6KAgPIb@E8o_6<6d31N~~Q YL(&lzSO5SZZ&jG)&9w0-I$#g~*~Fxq`2YX_ literal 0 HcmV?d00001 diff --git a/assets/images/our-models.png b/assets/images/our-models.png new file mode 100644 index 0000000000000000000000000000000000000000..c3845121acdc035d938dbebaa39b6e11343e748e GIT binary patch literal 21756 zcmb_^2Rzqp`}Wspp{+DXgF;F}Nr;AoWYZ){LiP$7EtQIr(KNDm_8tvc*<_Ch*=1(G z$3;Ew{a?@fzMs!~KlkG%}Ov&J)s2QSmTtHO42=hUsOizS^N`Zb?5`gtk3tGOjS>~`43g?0`}DF?VyPCS^A zX4s)Ae4sBU)y5^&(m{@IZ(-M(1^f;B6(1x$&x>@r5VdUCVRpym&t7PKkyKcb(b~O% zCZjR?y2nhis6^Du*fqV?iaJMQl(iw%bpEkuM-dGwejdRKFi$YR<7!id}{A4 z`qk0&i#9VW$sbv}Uo)=D|MZDcyZRi_@8g^%+~0X_KD|V<>gtmVqQBWE`@ZJOp1LmS zd!S^j?2^nUEDRBh6+om&iin}bCq^;gK3b4$k=(OEJ+t0*&; z)M52BDY$!Vhq=ji#d6JKg_5uG=6Ui8DD7GrD{&(9Aak=GZmi)Fpv|;kSpulKy zQAbA9$BFYDDN&cCPM+ka&+4f&%vA1gOgC@;vet=)a%ukkt5vrqCZNbmWV@c89s@1& zp6yc9_lgAlb6QufbT!RP3uIFc5>Grlwo1lCNBgd!-m`I8pE+|T^ju#>h{$>et4VIJkkV=^dJvFxwl=jAqX4c$Rl;H?`5btgP&M1O382 zv0u8Cq3%o{OB&N{xaU*CM~o8V;{%qF%N%ml2v$qCKIuv0AYn8yqJK+B%xHlV`(Qmy zbE$toKweRi^AU>P(&zfv#6)-V``a-wZ(mVjcDq`ZEUUk{qBPCQ^u@`o+1c6hOjPc@ zp&>K(?W%u%VXdR@{Kf0uc?nU=>S=BXUF>|`v4k4)eERf#VjwN9`2E&|{q!LFnRDk9 z@w*a6+Mn1qCed-em3+1}CMKq==#3D$)nnGy2e4P_ZzhnC7vxfRfJxZd(I~)_4nHZC@3VPH{4ZuHseNeqWJhXdJFOEicpV-Q!JfpXs6{o;{N0L7d;7{T)&2}W z2^AHUcENs%I!@MI?v}^S3)W2&JW9c0H>u{Gb4&B-y0vxJ8_UK#(|jhDKw7a=e^xAg z`0F|Aks58&slB0R<9Vb=8TO^#XzP~lE_MHaZJnk)5ioIl(JLF(# zB@v>_w)Cz3G1OUpHulQmyYJ+R4^4#Ud-Czw*i7U!s@}Mf*3#O_t-I);X*<)@Vy;ik z23q2AloNyD!yVPwCGJ~yYf~+wLRF)ttG5@>FIZq2E?pg}9LBt2#gR=8M=q{$Jh^q{ zU20*UL)+EbIBs#u!RE4-*6x-i%$ux{Q@EICmy4q07-m<8h7#k>x=&6HMFmi^fkw?m zJtwl%-J2nM5Wp1WM*Bbi8agsySSNO#j?EQA42v6Rx z2m9(0WuwwZUg~}g=U%X2fpYk1c?5ZR0bgc#vS~-xrw1zormKa7ggQcF-QC^m>+37t zX&k+K_pZ*&^pqJ-7VYfQ%FlnyDyVRK!BQ2^ofpi0lx*VMwoR|}t@uHsW`}kAbp6e{ zzlxZ#i&zgm(OAQojL{yQ(EO^pVV~C1xHDRwFwUl$R@BNHHrN>8;zRTfw#COw2J6 zn-MpSHD)`>Ig#A&MsaE&maQ5av7~|r4*bAkXAeB>{&1JEt-1M0l)heNXn;z@**2>t zlmbih-dbffH8pXyRB{ON-MDotf=g|ixK5utr&ph3jI|nEY1|WZ^2OUXLY7_LGm}9o zk>^9aL~P3P99Cp0t@XM3D`Cy*w(;Ta6sy5NguE&*-}KnxURUNdYl6Rhyw{N;vTvWV zx7~OM*4QcFIJ^C1G`548m31C?QoIUF_nqHfVZ{W#!6CjRwVDA|_^LN@8LTR_l&dMPATaX%K>o;XI2($Sfxy!JJ=_uPlvnMz?&g`O*xw$7KT^~Pw%=Gbzx3@PEUfh+!mlKEX zF|zEu`sMlVuBs^S6w7|E){NF1mxAJAS<%U#x4(v;@gqMkdM)_!fU)L-$k&={9H)P| z*fZB%P36E&pK|@)*Qe%V>L158F`4fb==SufufIQs8Z$PXBp``HkJyA%GA9Qf;>obS z94ss>R@T-$d-f>c)^g(;!q3DqGc)@Sv=tQO(}8+*be4m;Rf3a z3(RK#cp|o|MsgiIcotjnEH*Y44;hvHkfj@o%BeD1_)38MsjEClHd=JV&so>~>!lek z!?%;`x^+lxekEY2a{c;s=iYj#V=Xg4}C z;xT0Jm+8v9Y}rP2b#-#)^78W9>w;un1^)c`bD}6d!)_|2HqkIj+uK%UyGr)%CK~66rwF^3dw^ zx}V=Znzw#lJUH*xqD6~p-8ty?7d(th6J=nV3<j$-7* z4O`n)%a=dBl=^tRKzsGp9Xn*|67-+lzI`;Zlgs79=g-Fht|}2M`!0QEia4LFaMY10 z&$KQ$m}_csu=s4IU3yPX&xWL?9&WK3#VawKoSdCMO8v6c`omPCbUVueI9IH&y>Wxz zZu0un&yP#b#$A>7s$&(gRvPRmRvqnYP!2lvM!~jc@DgeeDgA2JodNfDK10QQ&&H`c z@I5c|Ol;uh&70$lTaO0>26laUv6q37v3zW>qwu|mP3q5|M*7w74u^zOv)#olFOFJ@MU39Uuii*chwY1>j?yBt@H;O3+ovQ4P%8XpK zUByU)l|lh3Xm4u^Kzifm;5dfL?U9!6P zssFuGgGp5g>xhO_%P^eNuCEcVgoTBNYW1RYem1`kRnN=-ExEksMW2xOxZVD1)fbmM z(k}K=MoAtUov{6=J+k5nf|PT-?fA28vMT4#zXoy$%E}T0A}8^E_T9ysWNCngaM-!4 z1QlwX(~)WrUYwHMRZ*UKmr2&Wfmkr8KOi#rWp8b!{md3y#+|@sPq}yRe!8!M_Y1DM zYQ>5V0~WFlJvwYP>WA*MByOqXH2Z>fJ7jpZc_z=YaNDdS?i{QxD+>hxj zLhZvVSFIBF1zro$Ji18ktC2ol!SxUY2LD|+`}(=B1o!>#-^Cw+jsMeEz-jz5`xQHd zKhAyiak$2HvyEZNlE8&p=WpB)&PloRM$pu8Get?wJ6^xz3C?wjdH3;6`ANv9-6S`b zdru$REhUFq+-XQTDBN6b$v9M@D62ElQ^WlBk@TrkD#-A7?6)YFipwbKrx-vI=ck5- z3)m}FB_-zU+55i!5x|~c&>*!N zzxkf+ozl*oJG&ZE6A}3$psNHxbFLS)OS7Bk70_=?d!dP!0se;o!&U^zvLA22Yw&Ak zt)Cy41AxY9e4PL1*8*Y^6IGF;_Zhbw`+FsujyINW&z}baV1jNU?=9eUkBEq9yxzJv zeP()s84C-lDZP5_+7Q%4+v%~6Ki_)T>!zZ@uy}E$zvQ#O4|ENQKuS_l{@OL(KaZ^d z1f4$C{_3T$)t68bOUP(`hgHb>4pY@CI#5!#J<6dc#wHk$)$qrGiOP99j>2(Vy za;rwNlJkB`$ia04MXMV**droB-|Jh$E~cGw^T@Adt*8i5lsK1Y;GUkIe(joxT8t~e zsq2%uJ#so5ud9Q20}3fx)#$<|bo|e%F{7#|A{T|HP|e7fCbpc)#d*8RJbJsYK3NePr^HS<+&FRaB=4%;p@9Jf zoE<}chs~Obu1gKp%$K0Jj%3bEr$TQC^*wrTb={@Nv+)G(4;lI_n^;Mm_+q>upC7DX z@!CCij?CSDCwR8_-;DRq?1zX&u9*9JO9+3&s{ZTiEI%ek4d0pjy0B0D%J@(xN)`SM zRZCA1PcR0JXy5i;eyMom*6N3apbzavpr@J3sE) zkm7=}P;hyVPMN=C_E!UK?GDzdjy`q?2?zI*oWl~xv&m6FPjPLGN*^s+bY zx7hkj5{M>HmX4BAH)6*jV)`ie(Q)yPgvY4w?iy>l$^sa*m6x-1>s3XDSoGF1Ffj1^ zvN)@7uFqoVhkQJO2;dMZ#$zB#W_I={Q0QP;??dYZO}^93*a6v$4;1i8`+JE_|Mueu zLC_;{IBF2BsWwLIA%_kQ$x{8yfZ;(Ycb>dmOvo~_u4c_yGawL}TbqZ()0__Wnd`Yt!?#OsxF+RmH%-psPCC z&!VT=xXG5>H#XK2C~hoa9^>lo)$LX4_R1$t+{ed4 zDT87>sJvjD-*z1(rDc?9(fbl|!w_UBCw3`1pPYQj-og1yqe?n(Ik>C^MN6z$>-*k~SrXvWpNxu#7ZJaVT` zKPR^WiUhH`qxK59pMAP+s22P5zVhw>Pl`$#Z~~|&VC9Xdy_&qd{47j$VfXgCFeb8L zEFwD{3tyO1w2MjR`L4sRL(>!euT*2W!cAKa>vLb7g573tO|MI-4&;@S0nl=uoJ43Cah;heO! zw+AAP5TK6y;t37G9B+L?v>iVihDu1N{o%uh!Dya>@^0R;r4or=cqULi({5mJ@EQK8 zP0$bX@!3eSkffyKJi9(D@~&NSZw{L0lSM9ixlbAw;8D*=143tBxzdkEIrJ(>2Xw(B zk+v>?TTt?$x>R^i4{d|)_x+{o&(G@(!8MPYn3yPdAF>?+$&`mc@b#U>$Dw-DVo;_t zWS0tG?nCVgJntfZI;T|wN?O3L)|~G6Dz$Rl{KNRg^CIl+oO}J89z9Y5eP6#en zx2j5Y<+f8_q6-1K-d@Ua*6FGUSzVsfuco|JMC8IYSuZ8rQ$@r%f%M6Ng5>s}o}ca1 zLKHY5K}msrldNO*J@@Gm(V!pYT2ZoeVj+xupz#EXyR{~O?uUWpL#Fj9@)qTjMC#A~ zwfjnfS;}=xVgJwz27` z)wVb%MC6cj|5*-6#bYnEowTWPN5R$gO2lvOs2b$`(rO%q{FjUMag9Sn62=QkSenJ*VO)uFr4P&u0s%Wbns*#($INc^TuS>RaXZ? z!E$nPQUK%|%4SiAG;oN=_0rk;skAe(S6&{nO(Ddwxw%<5sAV<@%Z1*?VJoz;x_EGRAAozwgE>(^v2Z4EY;1*El85g^TC zWnCsH^ZaH5s#i7oarXZSS^*hw0q@=tO zwHJ)^P&3H1x-Fj`fM_U#u%oJ?#c`$RHMeAP#^F6t0(3QgJdu%+z^f{e4`X=?Ki0mS zGLH20RECCvioy6Hjh znIHAZrn=Sdd;!COd{2VF=OM>~X~-j_3;C)?Cl1t>mdYm@HtnC8{<63DN2`md2S@r) znKZ;XR#`}G;>V7iK4}|M2eJM0`7IK2Le6a{ZR+9rJh{KFS4@rmC_O6S%v##vZSQqS zzX!x!2cdV@?$l}BDG4wpbRs*fn!s8)+ol(0#{KR~?Kg?b3?DMp*48G>4TpVJah`nu zq4y1teQ(cSTn354Y--$`NIZmRB6?S`vF+$A%1J@mK;N|@PN&=rf-drdZjLisvd6Om zg2cCP+H^Qb*86K;L#p|}cgc9&N-0-?SJlVX zt=q66a`F28T#)7f(MUqXDUSN_8eJMyW^b)h;4<%lg}07JiKn^uRplAPOPf7nJr{Hd;wM+bWYk{OcO@ zwJ(E@*!_Jb(h?(_ot@EQdV#0aQXAllvXvGTJ)6YG19vafM?JwsjQ^fllRP9 zaKVnGslJi=?Yz7vAs*+)4=l7D&+9F(nc|MJ8uCE5RysO5nyl4{(^=2f)zi3bn762p zaOVw0PAdq3`J?uXmUZ`;$M%oT%1Bvbv=qgG9+A)GT{A4N`6#Mc9<4UK`u~{*k+=9S z%99vPjobIzcu=P;l-ur9%l<%ny1Vt>l+UNzHYZLj1ab5G(FD$Xvw8XhBOezuvMxHU zEJzrHN7&9xjY3Cp*!+prl5%PLxKy+P+#aR-ulk^b^z3y1zMPVh8`|$1=cs4s``Zh` z&=Wy?|4V>?8rbV*BY>(trbw_^T+?phppBGirdk_)znYqJmCsBvJ-Z z1c;^sE-1Yp@(Q@X4_pymi8+a*)K5YWS}vi$f4&te(0KhQa%8zsf7;(0zN{E{vJ}L@ ztfzYSUp0oD8d^HKGaIl; zw*jBWDKu)U(8UCQ5I;O8?kH(!Xb^!LD{f;%BX(7)s#l`JJ7^aj)6&K;OZ6Y_LfVEvuiKbrB}hAgEK67) z13i(f90Jw|S%goIBxZIZWdI7fqdEql1Kx?8N%T4#jEmfAHof}*NQjDvMqU+&PN7~J z`vD^dS{m{KxiVCORjXGAs@qQ~v+O*d`zzoi0-V@2xL1(npg47)KhK(SCI_S?6|P*l z;ykf;?_L~phixwxbK(jl0Avr1eZN_E`N|bt$otHU)Xz__aELv4b-?)TahDBKCB5jc ziAT@o11?FT--+61z%0_2YJC!7zTH?Gi_<*vP-2o05FV1AzKx9y=%iPk_N7Z7^Y_fg z(dkG1(Q>50*PRe*1kJEJh?^wz}Qn|sQO+;EVP z9>l!)^V^>f+M$c@L^rpFXV6m}AMNLRmd7#cU_qysn`!QXTX}e-P!9O-S+*zBQC`Px z(5&=YNXjzAN3LUQf8}qFlA{-PCB>d-&a}|jxEAAQf%oq(WRTA%p-g}~5(|TQChi*T z&RFL4>!qI-JV#DYpeIHn{t&1)XpElPl=A0WPWGn#-ejh)>Nqt_F-kjkfTPom>8 zciJ3B_n$*mIs zos$!WB1nXiL4CFVQT??>;@ zW@Kz!`e5a@tHno&+6X}=*#M3&ceD(aupaP%`UqcFWOeNj*M z;SrNB$jz0zb}bwWvyFHz4jc%!n;Lg=b}lkKC}v_31L_k9Vc;y5m&kNTx$&3tRy3qp zDMFG#@1_hw3WSGkh`wiemp44UkZAB`Z2z-7f82B@%CDfo-WMkcG)cC#Y-)sDB%y10 zxYjhY1D3k~;5abtKZFw&@Z`xd1GB;mMo4ENSmv9=F>NLDbx4q==3TtyUgmHOQyh8a+c*$#&N^bL}p9W`acr!wCRPGa3nrVA zTk{wP>LuVxfDBRn5fx)g7sg{N$tFIR@s6ZQ5P?h&Pf&d0GM_B)eb;rciR1s;0A3TES5vX~@qGOec z;OK(tE^lTQhjZ|~rDe0MtgQI)P<}Bagrmeo1d_79 zDZ!%mtfT@^vZ;Qi)!-G#r=lw7P^b|2VeqwZ2@2NSm;k+C-mt-|l{QyG$MUe2XcJDH4>OR^yoT2!U~V9dg!`)wZo#1@@!MFwzlb={0N zNlr>)K4as!lsytZAl?yI|2uc+0iUGN@7pey%Gjk*T!mUt_J(`KXv*j4eq1u&PG>d%x(^GPar70ir{kMnHa{9!WH&(9B?bN%V5 z7usH>IqdeYqwF33tI@zJ+=3-eq^8&sIZe<&R$96oeuoU*8(_iZSZUFMBcM)VMJ89f4>Bv!TsgSP}Em0kc*CD@8ad3z3iv0I5{|o z&#^Y$Ch?S)a0s!8gu{?iUeXqMrl6J?Q1$Od&|UQ^^v%(Bs8l( zA;n}l(2P$s%|rsNh2l9#F+yhHl!SZp4InCSXOYAd(vTvg>G?}H@LF10Vi}Z3mndVR zPpC3j{$m=c?IDM6&nJiq!NSbOMp}9=tD@ClI^ck|zNsoW=HbJK1e1K+1p)eJj)CRj zx@S{2Y({!QHXU_*?%j_xFGHiuzYcr)`pirI z5Jig7Z7L!GkcgR%Ycp>GLzd>QOjfRRi0HiAw{IWcEq7o-{YBaWI?}Ako1#4WU_K<)T;g=4s4Um1@$RVckL=T8n+Qpv-VHUY`5j-(0G$go zgHy<|kM-{z#-JR7_}p%W4V?UJ0-_togFimk(=o)-hNcKaIcB=qdyqE&{ZpedMDZRm zlT*~iJQ&CPab7(B{JryiI#44sEte*YlCn;ybh(%J$S)LL)^VbSsIN|PckHHyRgc%- zZ28TOFrHm%sIwKF&s`1} zaPt-=et)^8@NxpCrge@&1gJvw8UpE+WAb^zDtuiI<@An)&zfC1Y*{P)&~%33Z^d1T zp%i;Q3?WHzAwkx6*Vpr@rJaCNcu5}=mRTn~#o%)9-aW`JT=a7Hr*S4iQFLb)UF_TZ zQOIl=4a>PW&u9GKRH3(VKCtQ&+f)M$2c2AGUl{kX_RJfWE?p`Khr8SfG?Kn| zbV$r@3G@4hUhb=asyVP3Zkd^)vpXmEn@1%)7oN{$=TpoiS@_gAi2s`6TF1L+5i^8k zi(J%x@SGqrQa~|y-@CT}F7r*l(T1snwQF{N@mQz#v?tE~o&wzM8=Byp4gLQ8G6;eK z6M1NQ%Pa+QAQ~GR2Zn~gT5&C>`HET&If?L0&6_t*es*od*Vm&Bz)deFMhszKP1yBq zP9%hUet@Bzanz7o=;N^8pq>krYo9^=b4?NvTnEvLb^Ro4i680L`s4p{b5|qHOIJ(| zFqfb&0tJ%q&$|@tVnRnBA_;RAax^I$=wHjt-l%wSL&F=)LS%t|SCZ)pEiL)EAGmOC zx_Jvo`{AwIx4G#_=a`sLo^PcicUS$ZHzdpN;p4|zFNp!$p{4EHx5GvYHgNKuKBEI+ z9(p4NGzdey0h9k1rM1n=!%s&aA>0s}>dConGbGs{H&a%0N(qhcx$XZ&c!ZP`!(D3N zTTAvsVv|<09af%Oli#;fHP+ahLPFr|R!Pm5 zK0>?H0=6}4ov@v8Y+l$;01jUE67Iybja-f$>`AF}YZCx?Cksy!`2cZcqoBu=4QAUw*= z%@-MHPcI~V!U`DkE0%#h1x{)qFYVV2V2d8Yx*>&rkeo#ay?_yz~d4n@v6>^*Xc#t3y~B2coscy?3u7Kvj5?^R}~)@w{wD&SVI#;hJ}i zi$?CcO89ACbek3LD`f3r!W!Yl&y@M4c7r&OAq3{P4uG75oLmLuL}8kT8pKcvseKhY zyX+<@8f+qI-g|j@%^RP<%YMLH$s9wE@EJTlwwn+IbP*sIug#{UXprHLfIROd74(Q3 zEqX)5g$5QIL_2Q=GJ<)31xtwl6a*FO-y^!xYk$GvC9*qvzHHj`rf zlOCAlF}eDz)y&5Yrgw-txj?=ozJRyqG7ZIB3I)RqV^gT-PMTD}GK{xS*J1nLm$4YSZj{(K4|gI**7%USh$sa+7azwUS!VHCQZK1pvrBU(N};PKm8$LGJz;Wn>_L?%;IkmxdE_sD=s6+n&L%o;tt9}FyxHc zIBAv~oaoQggKVZ-mk@!UaA5$1i20;L>Z`c@4%p!Q{OF`4XNVW7(+#7^rX%O{*@g#M4bw&T-z=5W65Z;ijC!hPk03D17 z?t-Pxpg!qv4i7Eef+b2uM(^OTg)tHkZriLI7ZP_q`U9jlf@wSq1YnZs%(bT61MsPm z5g15y_CHh;y@jn53OxBdqoSfn#~xB;E@~iwp~!T^A&K64)ahDb+i^2EG9+LUA(^B(cET@`~Ehr37`kGnOS%3r49 zPV@7`YTnaNI(pIca=U%|HpX@~ecT1>p;H_irV{Y2qZtVzw&QUeApmR1ePN10&cvGs zWmg9B2Y1ZK20@c!5Z4j3kD1r4vu?U5glB{CbL_!^J|en<3=c!`SHwyyK?Li(cy}W= zx8&X{UrLY@*bm-i66^ecLwv*-JwBod;qY!BDA~!~z@Yf zu(J*VX#54B?|?_DP#8SW zFM+b&kuYO-0Xu(%N>4hxWXh=N#^B>d+tCZ?7clFrhhq$Ms>J90r8n#Wok<-M(4sgc z(SbwC@5pZ6z2o#dF?10s0Oe;Zm0UFn1iU0GZ863N-Ax4+$dUdgK6`t62wSS~V--hV zfO=$=3L{J|+Pn}h+C!NB6`bn`fpQ&U-fOmgYxh-9XvrF}761B8A3 z@bxb=4uIFhs7J;!=3y9zd}`}Q*(1!J$;9+mbMU0%|W#pqla@^KW1Z-|7;hU^5V0>4w+xFd3=pH zFaoD15|7wjqwQUZTzy$z|Bc=&l+qKfq?_D20zSWbY{=~zf~}k|W}?KSNde%m1PQPL zv!;*N?mdoHC$wmbrwP6YY}cV&zx;sVqz=`QSBYRwyQL=A5EKfvg%}8DlR6iJj*bo_ zt*7bfQ*@sEH>xdw3S9e6h9yDG&~t9m*VpH4IHNdeKw3+1ReoVvw3UP7Dk?T6iPXLH z0hHyCSaElOKu!F`KVe4IX?ioB5q1 z%F!?uvyu4=47{pDH6et8KsqidzVsni4s51Oa^K+aurgc-&b?q%&o8}#5gk4;nI%h> zBnSElGoXdAnC+kvC}9Om0^c#cXVYvK8$@ZjSZhm5NtHnn zdKmE89u_7JNeCka=1m9`PDvb}?PdFa|Neayw*T>O_t@jEmcA8E6to@+ZZGzZ!dVX^ zdeV4OM-I**X(tlH$kGd2S6Y!4BZRIOn+eZizW47mFM$Q4J-uW+26YZyJ|6fsphJ~m z3_ZYr+r!+R$ zLrHFHyfG+4!XKs(GU83<00}e#Ttc2c>>MyD)ouK5Msk=7YGjj+Ei({TMTzS6Ysqfa8`{V_=Z$8U?e${OO2k&kLIeDSx zU4DYA%PQP5ysq5e{aJz0b;at~(v44U`z(w!`@Ln`TJbj~ZnsR71{^+V*xU1$N$2*< ztgOF~;k|bH;;5H-zaGm-%P+gWj#q|5#2|x~2+SNo$Y?-Og1}_zkmL0~1D)$a5{zgi zobc`2w_n4IH$WwE)F5ZlbvjM1v@&8E;|}PFaZFk9+v5=Fdr8Bk7Y8F!&9(jLSbcw4 zJZisV_;)J7Kf3Y6u>!gC3F;piQADc;JYpX|f9|eFORAJCIks&Jisx5RRSio{P9C1f zRBsz}7V47|7r%=+&Ugq7#JLT9XJ)L}o~Y@zqYaiR6Md5Xwkn^!Ujh9hJBbimMlYEhC}Pil~^L3V<@!2er9SMMx@AYhNP={ z_?>lVCnFda(f!!^=;h0GBQ3NBu3@S-)2!v!9r(p3`|(M(X7abGf?xd?E~u5)2+`K~ zL5U?qnoM;8-&pOX8t%MUNKrQdX}b@t@fnTQ-8*g_N!*8N3(^Du#^7M`S+5zwM5SuP8s{%Ad5ZqxnFtm9`+?#j z^4$hDpf5;{M9@#Uwt#~gxkuimdT>E}o81Svwi!e3qq6=WW|x-q@>d@hr;hx2|wE+da?Dq z?le@^l8Uci3E}{@L7(97J3Yn_IZiOFhN0-vtz0;F31UR`WAAvi9~X;BU(mw#|g2JRX>2CB9&s zLIM}lg0Yffs*9MIIPMr=81U`K339jr=b;~;NNlnb(lB=giM}`eA}k`J60&$O4!$m? zv(tR2>1}?2fsQhoG|gLWP;ijUmt${+f0W)9D{(BmQVtVcBup_#?hnN%Y(ldN2~>`d z;NbhaN{+*IMTVr<;6r%GOY=MD0ONe>prnOG0^!!^DSi+Oga24rS(#fvz_V5{E7_OeF z1^|>_Sm^HU&3@BtCCUpKl7^@@3}5ry79F@ndS*gGM12El;8?azd(VrZPb+1AKy_3- zb!y2?$_MJimIvs35CJ5tk_WvbT4PPPwL>E#Rm8>TX;8mM7tr0OsFbgGs_(ESqvS1j zTxru8;Aw`@3Aovo&nMwRgO$J1Wxg16<&cv5}j;%atx04T{jQ-Lvj};V*H{q#+}4^PTSh1qZw*e zEHZi(xSF2Ur#$*I8`YCrDbgs5vJI2W^80KyT_Hi=zkTz?y3MVunFK^9vSyN!4@lLc zjZR#8>X{_)zh~hUMLWBUYx}2L*O7Hg*@2IoAfID{gNi|{gEy<6VuS;EJ=ov>&fqc< z@RTbWVWjtrb}Wf{u^usK6uNqP%BN3rVTgp5rX=9YG31BtyDVxYKneTMIfoMmR#lBV zCJ(=$&4~CX!@KCk$!w%jSX1*U2fFLf7%;nW2$`!JrDEfgo#67q_*)p#4Ahpn4f|1z z9-6YE;+C*VsWlg_f~N3rbC(i3X@C0YiiHm69QXB6%^NNH>bYZU!tv%SkF$7;Ae+Dv zuTV=vZhz(r7Aq>!N}_|9qOn(QZ{F+xNS{l7lcs30kfkymLvx5U-iXBg=n)fbavu1$57I4lQ*2-OM%5GH82k#&isWZgkCF5gK;fN}(`GByVtx04&Go+WjMX&r8# z!##rsMj@F6ljLARhw`f|FYk-!wQ!lu-)3HjR?-6^hV`>&&o2Ca5F%O^Xo~9L!#C$I zVLgGXt=6lj$y#&+iw4_7ia-W|Y3;-O>=7r+8Vz+KO^fB(qK zOO19D`fD8zpLT_XhNzS-(L1N7{GOpIu)fHzyV>b*{SYRo2FKeA8WweL18ZkNzj>u22)$0TDq~W5d z_>5RJ%E}aiihW{Y6iE2aitdCs6tnsxWZ3)UNjKaS zvHgDj{Q2ty^qV(!-MPfb2%$(U@KeLVY*9PV)!*M_V0R_<41lo$Qh#4c_s&KRigkqqHs--KbZ7{i zFf}#3mz@?6AP)KJl;8|$NK{r<_O>@xRQQjNTLahL6<9>w%U+IUTS1K4M0Iy`oWD;% zz;g0o)2`P89LlrZRwMkH`GL>4_oat^`yd3ZK2k;pSg}Iec58lG$R=uc9-ktCSil0N zw;w4SvLLi|I&NO1hg_A+t3d$Zfp+mEm_h zlXYE_V4Y;>;>C-taU{>mo=EoyJ107yo}LRiU4W6g_fcUjWuysPUOZ`^AlWCLJJfn| zTgXnhqCnF1ElS4vMHM(YQU0o+Yb8tZkm8VKhpa?+g#F|%3BBZfuL{X9JDLEUC8l8^ zA*JvaYQ;_s8(-P~sKCwkhGU*nTcP-_Tr;*d1VoJzph@V6() z8^A~a?c6wq`x?%z+p`H2$$`G4`OCK8nA_V`SMb&;wjjyLxY2pIFo6zA}(&FrG8|bzi@$! z2Hm^S2XH^|>(@7N`VxI3Xxr560sQ-wAUtXo=l%IgHVhuve|#nR5FdXu`;}liiY&Fy zL)abRXM&9liM4U)*PAv_K_mxAObwbDXG4|@$WUqb(Wwd3sOljk&6Rntx8bM6P z$=MDoFwpBkNTj&^s)HlYz8IL8+;HhbnDrr2G%P3;2u^bhZ&Fjc(AcDPcNj7GOXWmX!T=dk`gj_Er^sGOt+h`yvR#^MNNgE2t?%8nfx4PT87 z(|$RUOogT`BuXM0L*pPI8`h$-KI16M8>%N{Wb{SycS;g5in#ER42Fh=?1v|Njc2z9 zXBzfed?HR<>eN$-4f5zcVNj~xx(@W76IoA2ckfq_FQ_wyqid>!GGYm%hSppjTYRXl zeI!hLx<9qx!1xUs5-X?tBO_-=17TkSswlON_Nh{{k(rxhQ8;?^_RjN3J_wZjhKBTp z*2&Qo&(^PPT6Jjp_a$N}1n`I$|$?yS+9bl@0 zwDiMk`(GjL!$kiI?@*_o7=jsEBFUkjR^4X|7+=>u(aqLWgY!EK)hi?|?fm{%5e?*{ z!u&IxQ-uC?)5*zafGmNL%h097S)|rFZjubew6DM+*h|XPrH@47Q^g)(7JWP_$1<>| z=Nf8|IwmKQ+oxt{U@^58SJd!yE|`rrV^k6!;-DfU^B6G^M5ZZFZ~zv^n{O|zeXI_b zGSNx_?g4eW@Rp>EDk)X$^SpPy|96oaEJ3i_TU?(xx>SS%A}-Q#Cy0m9Z_Es!&c1$> ztyv09l?W!7c+$ft9e^P56md!-=1XxVI3&p+M5S&eJ}(8Iz_N+B^We8e_#i%XuU>^8 z5$5CR#gF@$P+>e!8T0Z=F`t2t)hRLfLcsaQU1zHjCBW@qd)b8N61Xj|sCa41eCo$} zd`1;cmvvuq$7>5`2>DMq$mb!Ai3%@1y859wtQW0?i+u|I|K*suOoOQaWX1pZmGw24 zwvXQF2$dkOCfGtaB~cX8s;QDiguAcbO^J12Q#aFZ`$rxvX1 z_ka=qbEf-G`R>2F7sABx@d5w%icP{l|M*I<&fio!_ve2k+ Date: Tue, 28 Feb 2023 19:28:34 +0000 Subject: [PATCH 2/4] feat(post): complement model descriptions --- _posts/2023-02-28-our-models.md | 26 +++++++++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-) diff --git a/_posts/2023-02-28-our-models.md b/_posts/2023-02-28-our-models.md index 7dcf438..1dc23c9 100644 --- a/_posts/2023-02-28-our-models.md +++ b/_posts/2023-02-28-our-models.md @@ -15,6 +15,10 @@ We have packaged our models into a repository that makes it easy to use Tesselo's most common models. You can find the model references in our [Alquimodelia](https://github.com/tesselo/alquimodelia) repository. +The aim of Alquimodelia was to create an user friendly and easy way to use and +change parameters on the common model architectures used in Tesselo. No need of any +knowledge in keras or tensorflow, just some parameters and you had your model ready to use. + Depending on the context and the goal of the modeling, we have used a series of different models. They range from pixel based classifiers to time-series based U-Net type architectures. @@ -29,6 +33,10 @@ In our pre-processing pipeline we simply resampled all bands into the target res Usually this meant to upsample the lower resolution bands to the resolution of the band with the highest resolution. That is 10m for Sentinel-2 images for instance. +Or using the same approuch we would create super-resolution, by upsampling our imagery data +to the resolution of the target data. We had successful models that would build 1m resolution +images out of 10m resolution data. + ## Classifiers Here we are giving a quick overview of the different model types and their use cases. @@ -44,6 +52,22 @@ are detecting patterns in time series at different levels. 2D U-Net or ResNet based. +#### ResNet + +The ResNet architecture uses two-dimensional convolutions to provide a classification to +a given image. This has been used as a way to classify images with a single class. Or it +could be used to classify a single pixel, but with the context of the surroundings. + +#### 2D U-Net + +The 2D U-Net is similiar to ResNet in terms of the usage of two-dimensional convolutions, but +instead of giving one answer for each image, it responds with also an image. +Used in image classification and segmentation. + ### Time series of images -3D U-Net based. +#### 3D U-Net + +The 3D U-Net architeture follows the same patterns and the 2D, but instead of two-dimensional convolutions it uses three-dimensional convolutions, multiple images across time. +The answer would still be a single image, but produced with time context. +Great to surpass problems like clouds and other imagery artifacts. From 43b64ad5f46aca3fc0b473c16cb531fe5a397a4c Mon Sep 17 00:00:00 2001 From: Daniel Wiesmann Date: Wed, 1 Mar 2023 10:26:42 +0000 Subject: [PATCH 3/4] intermediate --- _posts/2023-02-28-our-models.md | 36 ++++++++++++++------------------- 1 file changed, 15 insertions(+), 21 deletions(-) diff --git a/_posts/2023-02-28-our-models.md b/_posts/2023-02-28-our-models.md index 1dc23c9..471cff5 100644 --- a/_posts/2023-02-28-our-models.md +++ b/_posts/2023-02-28-our-models.md @@ -8,16 +8,21 @@ description: "An introduction to Tesselo's AI modeling, explaining the model typ featured: false hidden: false --- -Tesselo's deep learning models are presented in this post. We have used them to -do large scale land cover modeling across the world. +Tesselo's most successful deep learning models are presented in this post. We have +used them to do large scale land cover modeling across the world. -We have packaged our models into a repository that makes it easy -to use Tesselo's most common models. You can find the model references -in our [Alquimodelia](https://github.com/tesselo/alquimodelia) repository. +We have packaged our most common models into a repository that makes it easy +to use them. You can find the model references in our +[Alquimodelia](https://github.com/tesselo/alquimodelia) repository. It contains +the detailed model definitions for our most successful models. We used Keras with a +Tensorflow backend for our modeling, so the definitions are written in that famework. -The aim of Alquimodelia was to create an user friendly and easy way to use and -change parameters on the common model architectures used in Tesselo. No need of any -knowledge in keras or tensorflow, just some parameters and you had your model ready to use. +The aim of Alquimodelia is to provide a user friendly way to use and change parameters +on the common model architectures used in Tesselo. The model classes can be created without +deep knowledge of keras or tensorflow. The main required parameters are the input and ouput +shape that the models will work with. Then, Arquimodelia will construct the models accordingly. + +## Model types Depending on the context and the goal of the modeling, we have used a series of different models. They range from pixel based classifiers to time-series based @@ -25,18 +30,6 @@ U-Net type architectures. ## Use all bands -For our modeling, we moslty used all available bands of the multispectral satellite -images. For Sentinel-2 we used the 10 bands that have 10m or 20m resolution. Similarly, -for Landsat we used the available bands. - -In our pre-processing pipeline we simply resampled all bands into the target resolution. -Usually this meant to upsample the lower resolution bands to the resolution of the -band with the highest resolution. That is 10m for Sentinel-2 images for instance. - -Or using the same approuch we would create super-resolution, by upsampling our imagery data -to the resolution of the target data. We had successful models that would build 1m resolution -images out of 10m resolution data. - ## Classifiers Here we are giving a quick overview of the different model types and their use cases. @@ -68,6 +61,7 @@ Used in image classification and segmentation. #### 3D U-Net -The 3D U-Net architeture follows the same patterns and the 2D, but instead of two-dimensional convolutions it uses three-dimensional convolutions, multiple images across time. +The 3D U-Net architeture follows the same patterns and the 2D, but instead of two-dimensional +convolutions it uses three-dimensional convolutions, multiple images across time. The answer would still be a single image, but produced with time context. Great to surpass problems like clouds and other imagery artifacts. From 480e76a7ac2aac259f9e7b208e79d71a9651363d Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jo=C3=A3o=20Santos?= Date: Mon, 3 Apr 2023 14:12:15 +0100 Subject: [PATCH 4/4] feat(post): multispectral image --- _posts/2023-02-28-our-models.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/_posts/2023-02-28-our-models.md b/_posts/2023-02-28-our-models.md index 471cff5..b442a72 100644 --- a/_posts/2023-02-28-our-models.md +++ b/_posts/2023-02-28-our-models.md @@ -30,6 +30,9 @@ U-Net type architectures. ## Use all bands +The multispectral imagery allows us to get context from the various ranges of light available. This could help overcome coud coverage over a site, or could easily identify Land Uses. +The usage of multiple bands allowed our models to freely aquire the best out of Open Data targeting the goal at hands. + ## Classifiers Here we are giving a quick overview of the different model types and their use cases.