Conference PaperPDF Available

Machine Learning Assisted Urban Filling

Authors:

Abstract

When drawing urban scale plans, designers should always define the position and the shape of each building. This process usually costs much time in the early design stage when the condition of a city has not been finally determined. Thus the designers spend a lot of time working forward and backward drawing sketches for different characteristics of cities. Meanwhile, machine learning, as a decision-making tool, has been widely used in many fields. Generative Adversarial Network (GAN) is a model frame in machine learning, specially designed to learn and generate image data. Therefore, this research aims to apply GAN in creating urban design plans, helping designers automatically generate the predicted details of buildings configuration with a given condition of cities. Through the machine learning of image pairs, the result shows the relationship between the site conditions (roads, green lands, and rivers) and the configuration of buildings. This automatic design tool can help release the heavy load of urban designers in the early design stage, quickly providing a preview of design solutions for urban design tasks. The analysis of different machine learning models trained by the data from different cities inspires urban designers with design strategies and features in distinct conditions.
0$&+,1( /($51,1* $66,67(' 85%$1 ),//,1*
-,$4, 6+(1 &+8$1 /,8 <8( 5(1DQG +$2 =+(1*
8QLYHUVLW\ RI 3HQQV\OYDQLD 3KLODGHOSKLD 86$
VMT#JPDLOFRP ]KKDR#GHVLJQXSHQQHGX
=KHMLDQJ 8QLYHUVLW\ RI 7HFKQRORJ\ +DQJ]KRX &KLQD
OLXFKXDQ#JPDLOFRP
8QLYHUVLW\ &ROOHJH /RQGRQ /RQGRQ 8.
\XHUHQ#XFODFXN
$EVWUDFW :KHQ GUDZLQJ XUEDQ VFDOH SODQV GHVLJQHUV VKRXOG DOZD\V
GHILQH WKH SRVLWLRQ DQG WKH VKDSH RI HDFK EXLOGLQJ 7KLV SURFHVV
XVXDOO\ FRVWV PXFK WLPH LQ WKH HDUO\ GHVLJQ VWDJH ZKHQ WKH FRQGLWLRQ
RI D FLW\ KDV QRW EHHQ ILQDOO\ GHWHUPLQHG 7KXV WKH GHVLJQHUV VSHQG
D ORW RI WLPH ZRUNLQJ IRUZDUG DQG EDFNZDUG GUDZLQJ VNHWFKHV IRU
GLIIHUHQW FKDUDFWHULVWLFV RI FLWLHV 0HDQZKLOH PDFKLQH OHDUQLQJ DV D
GHFLVLRQPDNLQJ WRRO KDV EHHQ ZLGHO\ XVHG LQ PDQ\ ILHOGV *HQHUDWLYH
$GYHUVDULDO 1HWZRUN *$1 LV D PRGHO IUDPH LQ PDFKLQH OHDUQLQJ
VSHFLDOO\ GHVLJQHG WR OHDUQ DQG JHQHUDWH LPDJH GDWD 7KHUHIRUH WKLV
UHVHDUFK DLPV WR DSSO\ *$1 LQ FUHDWLQJ XUEDQ GHVLJQ SODQV KHOSLQJ
GHVLJQHUV DXWRPDWLFDOO\ JHQHUDWH WKH SUHGLFWHG GHWDLOV RI EXLOGLQJV
FRQILJXUDWLRQ ZLWK D JLYHQ FRQGLWLRQ RI FLWLHV 7KURXJK WKH PDFKLQH
OHDUQLQJ RI LPDJH SDLUV WKH UHVXOW VKRZV WKH UHODWLRQVKLS EHWZHHQ WKH
VLWH FRQGLWLRQV URDGV JUHHQ ODQGV DQG ULYHUV DQG WKH FRQILJXUDWLRQ
RI EXLOGLQJV 7KLV DXWRPDWLF GHVLJQ WRRO FDQ KHOS UHOHDVH WKH KHDY\
ORDG RI XUEDQ GHVLJQHUV LQ WKH HDUO\ GHVLJQ VWDJH TXLFNO\ SURYLGLQJ D
SUHYLHZ RI GHVLJQ VROXWLRQV IRU XUEDQ GHVLJQ WDVNV 7KH DQDO\VLV RI
GLIIHUHQW PDFKLQH OHDUQLQJ PRGHOV WUDLQHG E\ WKH GDWD IURP GLIIHUHQW
FLWLHV LQVSLUHV XUEDQ GHVLJQHUV ZLWK GHVLJQ VWUDWHJLHV DQG IHDWXUHV LQ
GLVWLQFW FRQGLWLRQV
.H\ZRUGV $UWLILFLDO ,QWHOOLJHQFH 8UEDQ 'HVLJQ *HQHUDWLYH
$GYHUVDULDO 1HWZRUNV 0DFKLQH /HDUQLQJ
 ,QWURGXFWLRQ
 )81'$0(17$/ 0(&+$1,60
*HQHUDWLYH DGYHUVDULDO QHWZRUN *$1 LV D PRGHO IUDPHZRUN LQ PDFKLQH OHDUQLQJ
WKDW LV GHVLJQHG WR OHDUQ DQG JHQHUDWH LPDJH GDWD *RRGIHOORZ 3RXJHW$EDGLH
HW DO  ,Q *$1 D QHWZRUN JHQHUDWLRQ FDQGLGDWH LV JHQHUDWHG DQG WKHQ WKH
GLVFULPLQDWLQJ QHWZRUN HYDOXDWHV LW 7KH JDPH LV RUJDQL]HG E\ GDWD GLVWULEXWLRQ
7\SLFDOO\ WKH JHQHUDWHG QHWZRUN OHDUQLQJ PDSV IURP WKH SRWHQWLDO VSDFH WRWKH
GDWD GLVWULEXWLRQ RI LQWHUHVW DQG WKH GLVFULPLQDWLQJ QHWZRUN GLVWLQJXLVKHV WKH
FDQGLGDWHV JHQHUDWHG E\ WKH JHQHUDWRU IURP WKH UHDO GDWD GLVWULEXWLRQ :DQJ /LX
5( $QWKURSRFHQH 3URFHHGLQJV RI WKH WK ,QWHUQDWLRQDO &RQIHUHQFH RI WKH $VVRFLDWLRQ IRU &RPSXWHU$LGHG
$UFKLWHFWXUDO 'HVLJQ 5HVHDUFK LQ $VLD &$$'5,$  9ROXPH     DQG SXEOLVKHG E\ WKH
$VVRFLDWLRQ IRU &RPSXWHU$LGHG $UFKLWHFWXUDO 'HVLJQ 5HVHDUFK LQ $VLD &$$'5,$ +RQJ .RQJ
 - 6+(1 (7 $/
HW DO  ILJXUH  7KH WUDLQLQJ JRDO RI WKH JHQHUDWRU LV WR LQFUHDVH WKH HUURU
UDWH RI WKH GLVFULPLQDWRU IRU H[DPSOH WR ³GHFHLYH´ WKH GLVFULPLQDWRU E\ JHQHUDWLQJ
QHZ FDQGLGDWHV WKDW WKH GLVFULPLQDWRU FRQVLGHUV UHDO ,Q FRQWUDVW WKH JRDO RI WKH
GLVFULPLQDWRU LV WR LQFUHDVH WKH HUURU RI WKH GLVFULPLQDWRU
)LJXUH  7KH IUDPHZRUN RI *$1V E\ :DQJ /LX HW DO 
7KH WUDLQLQJ GDWDVHW LV XVHG DV WKH LQLWLDO WUDLQLQJ GDWD IRU WKH GLVFULPLQDWRU
7UDLQLQJ LQYROYHV SUHVHQWLQJ VDPSOHV RI WKH WUDLQLQJ GDWDVHW ZLWK WKHP XQWLO
DFFHSWDEOH DFFXUDF\ LV DFKLHYHG 7KH JHQHUDWRU WKHQ LV WUDLQHG ZLWK WKH VDPH
GDWDVHW WR JHQHUDWH WKH IDNH LPDJHV DV VLPLODU DV SRVVLEOH WR WKH UHDO LPDJH
DQG PDNH WKH GLVFULPLQDWRU PLVVLGHQWLI\ WKH UHDO DQG WKH IDNH LPDJHV 7KH
EDFNSURSDJDWLRQ DOJRULWKP XSGDWHV WKH QHWZRUN SDUDPHWHUV WUDLQLQJ WKH JHQHUDWRU
WR JHQHUDWH EHWWHU LPDJHV ZKLOH WKH GLVFULPLQDWRU FDQ PRUH FORVHO\ PDUN WKH
FRPSRVLWH LPDJHV 7KH JHQHUDWRU LV W\SLFDOO\ D FRPELQDWLRQ RI FRQYROXWLRQDO
QHXUDO QHWZRUNV DQG GHFRQYROXWLRQDO QHXUDO QHWZRUNV ZKLOH WKH GLVFULPLQDWRU LV
PDGH IURP FRQYROXWLRQDO QHXUDO QHWZRUNV
 /,7(5$785( 5(9,(:
%HVLGHV WKH DSSOLFDWLRQ LQ DUFKLWHFWXUDO DQG XUEDQ WDVNV *$1 FDQ EH XVHG WR
FUHDWH SKRWRV RI ILFWLRQDO IDVKLRQ ZLWKRXW WKH QHHG WR KLUH PRGHOV SKRWRJUDSKHUV
DUWLVWV RU VWXGLRV DQG YHKLFOHV .LP DQG &KR  ,W FDQ DOVR EH XVHG WR FUHDWH
IDVKLRQ FDPSDLJQV LQFOXGLQJ D PRUH GLYHUVH VHW RI PRGHOV ZKLFK PD\ LQFUHDVH
WKH SXUFKDVH LQWHQWLRQ RI SHRSOH &KHQJ DQG /LX  ,Q WKH VFLHQWLILF ILHOG
*$1 FDQ LPSURYH DVWURQRPLFDO LPDJHV DQG VLPXODWH JUDYLW\ OHQVHV IRU GDUN PDWWHU
UHVHDUFK 0LUDEDO &KDUOHV HW DO  5HVHDUFKHUV VXFFHVVIXOO\ VLPXODWHG WKH
GLVWULEXWLRQ RI GDUN PDWWHU LQ D SDUWLFXODU GLUHFWLRQ LQ VSDFH DQG SUHGLFWHG WKH
JUDYLWDWLRQDO OHQV WKDW ZLOO RFFXU $JDUZDO 'DYp HW DO  *$1V KDYH
DOVR EHHQ SURSRVHG DV D IDVW DQG DFFXUDWH PHWKRG IRU JHQHUDWLQJ VLPXODWLRQVLQ
FDORULPHWHUV JUDQXOH VKRZHU DQG KLJK HQHUJ\ SK\VLFV H[SHULPHQWV 3DJDQLQL GH
2OLYHLUD HW DO 
,Q GHVLJQ GRPDLQ 6WHLQIHOG 6WHLQIHOG 3DUN HW DO  LQJHQLRXVO\
WUDQVIRUPV ' DUFKLWHFWXUDO IRUPV LQWR ' WKUHHYLHZ GDWD WKHQ XVHV ' &11
QHXUDO QHWZRUNV DV D GHVLJQ HYDOXDWRU DQG XVHV JHQHWLF DOJRULWKPV WR ILQG RXW WKH
FXUUHQW HYDOXDWLRQ V\VWHP VXFK DV WKH PRVW WKH RSWLPDO KRXVLQJ IRUP VROXWLRQ
OLNH EDQDQD GHVLJQ 'LIIHUHQW IURP GLUHFWO\ JLYLQJ WKH PRGHO WR '&11 OHDUQLQJ
DOWKRXJK WKH GDWD WUDQVODWLRQ PHWKRG FDQQRW GHVFULEH DOO WKH IRUPV LW LPSURYHV WKH
0$&+,1( /($51,1* $66,67(' 85%$1 ),//,1* 
DFFXUDF\ RI WKH VKDSHG VWRUDJH DQG VSHHGV XS WKH DOJRULWKP $OVR =KHQJ =KHQJ
 XVHV WKH WUDQVIRUPDWLRQ RI FLW\ PDSV WR VDWHOOLWH LPDJHV +XDQJ +XDQJ DQG
=KHQJ  JHQHUDWHV WKH IXUQLWXUH OD\RXW E\ *$1 =KHQJ =KHQJ DQG +XDQJ
 UHFRJQL]HV GLIIHUHQW URRPV LQ DUFKLWHFWXUDO SODQV $QG .YRFKLFN .YRFKLFN
 DSSOLHV *$1 WR UHFRJQL]H GLIIHUHQW DUFKLWHFWXUDO HOHPHQWV GRRUV ZLQGRZV
HWF
 352-(&7 *2$/
)RU WKH SUHYLRXV UHVHDUFK WKH LPDJHWRLPDJH QHWZRUN RQO\ GHDOV ZLWK
DUFKLWHFWXUDO GHVLJQ SUREOHPV VXFK DV WKH JHQHUDWLRQ RI SODQ GUDZLQJV DQG IRUPV
+RZHYHU XUEDQ GHVLJQ LV DQRWKHU ZLGHO\ GLVFXVVHG GRPDLQ WKDW KLJKO\ GHSHQGV
RQ WKH LPDJHEDVHG GDWD ,Q XUEDQ GHVLJQ GHVLJQHUV XVH VLPSOLILHG FLW\ PDSV WR
H[SUHVV WKHLU GHVLJQ LGHDV WKXV SURYLGLQJ D SRVVLELOLW\ WR DSSO\ WKH LPDJHEDVHG
QHXUDO QHWZRUN WR DVVLVW WKH GHVLJQ SURFHVV
7KXV LQ WKH DJH RI $UWLILFLDO ,QWHOOLJHQFH ZH SURSRVH D 0DFKLQH /HDUQLQJ
PHWKRG WR WHDFK WKH FRPSXWHU WR SURYLGH WKH XUEDQ SODQ GUDZLQJV LQ WKH HDUO\
GHVLJQ VWDJH UHOHDVLQJ WKH KHDY\ ORDG IURP WKH XUEDQ GHVLJQHU *HQHUDWLYH
$GYHUVDULDO 1HWZRUN *$1 VXSSRUWV WKH OHDUQLQJ DQG JHQHUDWLRQ RI LPDJHV RI
XUEDQ GHVLJQ SODQV 7KH ILQDO JRDO RI WKLV UHVHDUFK LV WR EXLOG D WRRO WKDW WDNHV
LPDJHV VKRZLQJ WKH VLWH FRQGLWLRQV URDGV JUHHQ ODQGV ULYHUV RI D FLW\ DVWKH
LQSXW DQG RXWSXWV LPDJHV VKRZLQJ WKH XUEDQ GHVLJQ SODQV EXLOGLQJV
 0HWKRGRORJ\
 /$%(//,1* 58/(6
,Q XUEDQ GHVLJQ FLW\ PDSV DUH XVXDOO\ SUHVHQWHG DV FRORUHG LPDJHV IRU
VLPSOLILFDWLRQ LQ ZKLFK GLIIHUHQW FRORUV UHSUHVHQW YDULRXV HOHPHQWV LQ WKH PDSV
,Q WKLV UHVHDUFK ILYH FRORUV ZHUH VHOHFWHG WR UHSUHVHQW WKRVH HOHPHQWV WKDW LV
UHG 5*% FRUUHVSRQGV WR WKH URDG JUHHQ 5*% WR JUHHQ ODQG EOXH
5*% WR WKH ULYHU ZKLWH 5*% WR WKH EXLOGLQJ DQG EODFN
5*% WR WKH HPSW\ SL[HOV ILJXUH  7KHVH ILYH FRORUV DV ZHOO DV WKHLU
FRUUHVSRQGLQJ GHVLJQ HOHPHQWV FDQ UHSUHVHQW PRVW RI WKH VLWXDWLRQV LQ XUEDQ
GHVLJQ WDVNV
 - 6+(1 (7 $/
)LJXUH  &RORXUHG ODEHOOHG PDS UHG URDG JUHHQ JUHHQ ODQG EOXH ULYHU ZKLWH EXLOGLQJ
EODFN HPSW\ D FRPPRQ FLW\ PDS ZLWK DOO HOHPHQWV E FRQGHQVH FLW\ PDS ZLWK RQO\ URDGV
DQG EXLOGLQJV F ORRVH FLW\ PDS ZLWK ODUJH JUHHQ ODQGV DQG ULYHUV
 '$7$ &2//(&7,1*
7R WUDLQ WKH QHWZRUNV WKH GDWDVHW RI GLIIHUHQW FLW\ PDSV ZHUH FROOHFWHG ILJXUH
 (LJKW FLWLHV ZHUH VHOHFWHG DQG FODVVLILHG LQWR IRXU WLHUV LQ ZKLFK WKH ILUVWWLHU
FLWLHV FRQWDLQ %HLMLQJ 6KDQJKDL DQG 6KHQ]KHQ 7KHQ WKH VHFRQGWLHU FLWLHV DUH
&KHQJGX )X]KRX DQG *XL\DQJ )RU WKH WKLUGWLHU FLW\ DQG WKH IRXUWKWLHU FLW\
.XQPLQJ DQG .DLIHQJ ZHUH VHOHFWHG 7KRVH HLJKW FLWLHV FDQ UHSUHVHQW WKH W\SLFDO
XUEDQ GHVLJQ SURMHFWV LQ &KLQD ZLWK GLIIHUHQW VFDOHV DQG WKH\ KDYH WKH YDOXH WR EH
FRPSDUHG DQG OHDUQHG E\ *$1 LQ PDFKLQH OHDUQLQJ
)LJXUH  &ROOHFWHG FLW\ PDSV D 6KHQ]KHQ E .XQPLQJ F *XL\DQJ G &KHQJGX H
6KDQJKDL I .DLIHQJ J %HLMLQJ K )X]KRX
0$&+,1( /($51,1* $66,67(' 85%$1 ),//,1* 
 1(7:25. 75$,1,1*
:LWK WKH FROOHFWHG GDWD GLIIHUHQW QHXUDO QHWZRUNV ZHUH WUDLQHG $V PHQWLRQHG
EHIRUH WKH LQSXW GDWD FRQWDLQV WKH UHTXLUHG LQIRUPDWLRQ WR VWDUW DQ XUEDQ GHVLJQ
SURMHFW WKDW LV WKH HOHPHQWV RI URDGV JUHHQ ODQGV DQG ULYHUV $QG WKH RXWSXW GDWD
UHYHDOV WKH GHVLJQ FRQWHQWV RI WKH SURMHFW WKDW LV WKH GHVLJQHG EXLOGLQJVDVZHOO
DV WKH HQYLURQPHQWV 7KXV WKH LPDJHV RI WKH UHTXLUHG LQIRUPDWLRQ DQG WKH LPDJHV
FRQWDLQLQJ EXLOGLQJV RI WKH FLW\ DUH IHG LQWR WKH PDFKLQH OHDUQLQJ QHXUDO QHWZRUN
VHSDUDWHO\ ILJXUH  $OVR WKH ODUJH PDS RI WKH ZKROH FLW\ LV GLYLGHG LQWR VPDOO
VDPSOHV RI  SL[HOV ZKLOH PDWFKLQJ WKH GDWD VWUXFWXUH RI *$1
)LJXUH  D WKH LQSXW LPDJH E WKH RXWSXW LPDJH
7DNLQJ WKH WUDLQLQJ RI %HLMLQJ FLW\ PDS DV DQ H[DPSOH ILJXUH  VKRZV WKH
WUDLQLQJ ORVV RI WKH JHQHUDWRU DQG WKH GLVFULPLQDWRU 0HQWLRQHG EHIRUH GXULQJ
WKH WUDLQLQJ SURFHVV WKH JHQHUDWRU DQG WKH GLVFULPLQDWRU FRPSHWH ZLWK HDFK RWKHU
WR LPSURYH WKH SHUIRUPDQFH 7KHUHIRUH WKH QRQFRQYHUJHQFH RI WKH ORVV YDOXHV
VKRZV WKH IDFW WKDW WKH JHQHUDWRU DQG WKH GLVFULPLQDWRU DUH HYROYLQJ ZLWK HDFK
RWKHU WR DFKLHYH D VPDOOHU ORVV YDOXH ZKLOH EHLQJ FKHDWHG E\ WKH RWKHU WR REWDLQ
ODUJHU ORVV YDOXHV WKXV WKH WUDLQLQJ LV VXFFHVVIXO
)LJXUH  7UDLQLQJ ORVV YDOXHV RI WKH *HQHUDWRU DQG WKH 'LVFULPLQDWRU
 - 6+(1 (7 $/
 5HVXOWV
 75$,1,1* '$7$ $&&85$&<
$IWHU WUDLQLQJ WKH QHXUDO QHWZRUNV FDQ EH XVHG WR JHQHUDWH QHZ PDSV RI WKH FLW\
EDVHG RQ WKH LQSXW LPDJHV %DVHG RQ WKH UHVXOWV WKH RULJLQDO LQSXW LPDJHV DUH
ILUVWO\ EH IHG LQWR WKH QHXUDO QHWZRUN DQG WKH JHQHUDWHG LPDJHV DUH FRPSDUHG ZLWK
WKH RULJLQDO RXWSXW LPDJH WR HYDOXDWH WKH VLPLODULW\ ,I WKH JHQHUDWHG LPDJHV VKDUH
D VLJQLILFDQW UHVHPEODQFH WR WKH DXWKHQWLF RXWSXW LPDJHV WKH XUEDQ SODQQLQJ VW\OH
FDQ EH HYDOXDWHG DV FKDUDFWHULVWLF ZKLFK LV HDV\ WR OHDUQ ,I QRW XUEDQ SODQQLQJ LV
XQFKDUDFWHULVWLF ZKLFK LV KDUG WR SURFHVV
%DVHG RQ WKH DVVXPSWLRQ DERYH JHQHUDOO\ VSHDNLQJ WKH DFFXUDFLHV RI VRPH
PRGHOV DUH LGHDO ZKLOH VRPH DUH QRW )RU LQVWDQFH IURP WKH VHW RI WKH PDSV E\
HYDOXDWLQJ )X]KRX FLW\ WKH JHQHUDWHG LPDJHV DUH VLPLODU WR WKH RULJLQDO LPDJHV
RI WKH FLW\ ILJXUH  7KXV LW FDQ EH FRQFOXGHG WKDW WKH XUEDQ SODQQLQJ VW\OH
RI )X]KRX FLW\ LV ZHOOOHDUQHG E\ WKH V\VWHP ZKLFK PHDQV WKH WUDLQLQJ GDWD KDV
KLJK DFFXUDF\ )URP WKH SHUVSHFWLYH RI XUEDQ SODQQLQJ WKLV UHVXOW UHYHDOV WKDW WKH
GHVLJQ VWUDWHJLHV LQ )X]KRX FLW\ DUH PRUH XQFRPSOLFDWHG DQG PRUH XQLIRUPHG
)LJXUH  5HVXOW FRPSDULVRQ RI )X]KRX FLW\ JHQHUDWHG E\ )X]KRX PRGHO
+RZHYHU LQ DQRWKHU FDVH E\ HYDOXDWLQJ WKH FLW\ PDSV RI %HLMLQJ FLW\ WKH
JHQHUDWHG LPDJHV RI WKH FLW\ KDYH FRPSDUDWLYHO\ OHVV VLPLODULW\ ZLWK WKH RULJLQDO
RQHV ILJXUH  )RU H[DPSOH WKH JHQHUDWHG EXLOGLQJV DQG WKH UHDO EXLOGLQJV LQ
WKH DUHDV ZLWK D \HOORZ ER[ PDUNHG DUH YHU\ GLIIHUHQW WKH GLIIHUHQFH H[FHHGV
0$&+,1( /($51,1* $66,67(' 85%$1 ),//,1* 
WKH DFFHSWDEOH UDQJH 7KHUHIRUH WKH XUEDQ SODQQLQJ VW\OH RI %HLMLQJ FLW\ LV QRW
ZHOOPDQLIHVWHG RQ WKH JHQHUDWHG LPDJH 7KXV LW FDQ EH FRQFOXGHG WKDW WKH
OHDUQLQJ UHVXOW KDV ORZHU DFFXUDF\ DQG WKH GHVLJQ VWUDWHJLHV RI WKH FLW\ LQ GLIIHUHQW
DUHDV DUH PRUH GLYHUVH LQ ODUJHU FLWLHV IRU H[DPSOH %HLMLQJ
)LJXUH  5HVXOW FRPSDULVRQ RI %HLMLQJ FLW\ JHQHUDWHG E\ %HLMLQJ PRGHO
 7(67,1* '$7$ $&&85$&<
$ORQJ WKH SURFHVV FURVVYDOLGDWLRQ LV DOVR XWLOL]HG IRU WKH HYDOXDWLRQ )LUVWO\
WUDLQHG E\ HLJKW GLIIHUHQW JURXSV RI FLW\ PDSV  *$1 PRGHOV UHSUHVHQWLQJ HLJKW
FLWLHV DUH REWDLQHG 7KHQ E\ IHHGLQJ DOO LQSXW LPDJHV RI WKH HLJKW FLWLHV WRWKH
HLJKW PRGHOV  JURXSV RI RXWSXW LPDJHV DUH JHQHUDWHG UHSUHVHQWLQJ WKH SUHGLFWHG
XUEDQ GHVLJQ VROXWLRQV EDVHG RQ WKH FRUUHVSRQGLQJ LQSXW VLWH FRQGLWLRQV DQG WKH
WUDLQHG *$1 PRGHOV )LJXUH  VKRZV WKH VHOHFWHG UHVXOWV RI WKLV FURVVYDOLGDWLRQ
H[SHULPHQW )RXU FLWLHV IURP HDFK WLHU 6KDQJKDL &KHQJGX .XQPLQJ DQG
.DLIHQJ DUH FKRVHQ DQG WKHLU FURVVYDOLGDWLRQ UHVXOWV VKRZ WKH JHQHUDWHG XUEDQ
SODQQLQJ VROXWLRQV E\ PDFKLQH OHDUQLQJ ZLWK GLIIHUHQW GHVLJQ VFDOHV )RU H[DPSOH
WKH JHQHUDWHG SODQV XVLQJ WKH VLWH FRQGLWLRQV RI &KHQJGX DQG WKH PRGHO RI 6KDQJKDL
ILJXUH  E FDQ EH UHJDUGHG DV WKH XUEDQ SODQQLQJ VROXWLRQ ZLWK WKH GHVLJQ
VWUDWHJLHV IURP 6KDQJKDL EXW LQ WKH DUHD RI &KHQJGX
7R EH VSHFLILF FLW\ PDSV RI WKH ILUVW WLHUV DUH XVHG WR JHQHUDWH FLW\ PDSV RI
WKH VHFRQG WKLUG DQG IRXUWKWLHUV EDVHG RQ WKH LQSXW VLWH FRQGLWLRQ LPDJHV RI WKH
 - 6+(1 (7 $/
FLWLHV ILJXUH  E F G 7KH UHVXOW RI JHQHUDWHG LPDJHV FDQ EH FRPSDUHG WR
WKH RULJLQDO VHW RI FLW\ PDSV RI WKH VHFRQG WKLUG DQG IRXUWKWLHU FLWLHV ILJXUH 
E F G ,Q WKLV ZD\ LW LV HDVLHU WR WHOO WKH GLVWLQFWLRQ EHWZHHQ WKH RULJLQDO
DQG JHQHUDWHG LPDJHV ZKLFK LV FRQFOXGHG DV WHVWLQJ GDWD DFFXUDF\ $OVR WKH XUEDQ
SODQQLQJ VW\OH RI IRXU WLHUV LV VKRZQ RQ WKH JHQHUDWHG PDS ZKLFK FDQ EH FRPSDUHG
WR WKH RULJLQDO XUEDQ WH[WXUHV RI WKH VW\OH LPDJH WR UXOH RXW WKH DFFXUDF\
)LJXUH  *HQHUDWHG UHVXOWV E\ GLIIHUHQW *$1 PRGHOV
,Q GHWDLO DFFRUGLQJ WR WKH UHVXOWV VKRZQ LQ ILJXUH  WKH GHVLJQ VWUDWHJLHVLQ
HDFK FLW\ DUH GLIIHUHQW %DVHG RQ WKH JHQHUDWHG LPDJHV LQ HDFK FROXPQ WKH\ DUH
JHQHUDWHG E\ WKH VDPH PRGHO WUDLQHG IURP WKH VDPH GDWDVHW EXW ZLWK GLIIHUHQW
LQSXWWHG VLWH FRQGLWLRQ LPDJHV 7KH PRGHOV LQ WKH ILUVW DQG VHFRQGWLHU FLWLHV
DOZD\V JHQHUDWH D ODUJH QXPEHU RI EXLOGLQJV IRU H[DPSOH LQ ILJXUH G DQG G
WKH JHQHUDWHG XUEDQ SODQQLQJ LPDJHV DUH XQDEOH WR EH XQGHUVWRRG VLQFH WKH URDG
V\VWHP LV YHU\ ORRVH LQ WKH IRXUWK WLHU FLW\ LW¶V KDUGO\ VHHQ LQ WKH GDWDVHW RI WKH ILUVW
DQG VHFRQG FLWLHV WKXV WKH PRGHOV SHUIRUP SRRUO\ %HVLGHV WKH PRGHOV WUDLQHG E\
WKH FLW\ LPDJHV RI WKH WKLUG DQG IRXUWKWLHU FLWLHV DOZD\V JHQHUDWH IHZHU EXLOGLQJV
)RU H[DPSOH WKH JHQHUDWHG LPDJHV LQ ILJXUH D DQG D DUH YHU\ GLIIHUHQW WKDQ
WKH JURXQG WUXWK LPDJH LQ ILJXUH D WKH EXLOGLQJ DUHDV DUH IDU OHVV WKDQ WKH UHDO
0$&+,1( /($51,1* $66,67(' 85%$1 ),//,1* 
VLWXDWLRQ 7KHUHIRUH WKH PRVW GLYHUVH GHVLJQ VWUDWHJ\ LQ XUEDQ GHVLJQ LQ &KLQD LV
WKH QXPEHU RI EXLOGLQJV WKDW HDFK FLW\ ZLOO KDYH
)XUWKHUPRUH EDVHG RQ WKH JHQHUDWHG LPDJHV LQ HDFK URZ WKH VLWH FRQGLWLRQ
LPDJH LV LQSXWWHG WR GLIIHUHQW *$1 PRGHOV 7KH VDPH FRQFOXVLRQ RI WKH EXLOGLQJ
GHQVLW\ FDQ EH UHDFKHG E\ YLVXDOO\ REVHUYLQJ DQG FRPSDULQJ WKH UHVXOWV +RZHYHU
WKH VSHFLILF GHVLJQ VW\OHV RI HDFK FLW\ FDQ DOVR EH DQDO\]HG )RU H[DPSOH E\
FRPSDULQJ ILJXUH D D D DQG D ZH FDQ VHH WKDW WKH EXLOGLQJV JHQHUDWHG
E\ WKH 6KDQJKDL PRGHO DQG WKH .DLIHQJ PRGHO DUH SODFHG PRUH UHODWHG WR WKH URDG
V\VWHP ZKLOH WKH EXLOGLQJV JHQHUDWHG E\ WKH &KHQJGX PRGHO DQG WKH .XQPLQJ
PRGHO DUH SODFHG PRUH LQGHSHQGHQWO\ $OVR E\ FRPSDULQJ ILJXUHV F F F
DQG F WKH EXLOGLQJV LQ WKH .XQPLQJ PRGHO DQG WKH .DLIHQJ PRGHO DUH PRUH
VLJQLILFDQW DV IHZHU VLQJOH XQLWV ,Q FRQWUDVW WKH EXLOGLQJV LQ WKH 6KDQJKDL PRGHO
DQG WKH &KHQJGX PRGHO DUH PRUH VHSDUDWHG DV VPDOOHU XQLWV 7KHVH SKHQRPHQD
UHYHDO WKH VSHFLILF GHVLJQ VWUDWHJLHV RI HDFK FLW\ EDVHG RQ WKH XUEDQ SODQQLQJ GDWD
 &RQFOXVLRQ
*HQHUDOO\ VSHDNLQJ E\ FRPSDULQJ WKH JHQHUDWHG LPDJHV RI FLW\ PDSV WR WKH
RULJLQDO FRXQWHUSDUW WKH WZR VHWV RI LPDJHV FDQ LQVSLUH WKH XUEDQ UHVHDUFKHUV ZLWK
GLIIHUHQW RU VLPLODU XUEDQ SODQQLQJ VW\OHV 7KHUHIRUH LW FDQ EH FRQFOXGHG WKDW WKH
OHDUQLQJ SURFHVV RI *$1 XVLQJ XUEDQ SODQQLQJ GDWDVHW LV WKULYLQJ DQG KDV KLJK
DFFXUDF\
%\ FRQWLQXDOO\ OHDUQLQJ WKH LPDJHV RI DOO NLQGV RI FLWLHV *$1 LV FDSDEOH RI
FRQGXFWLQJ XUEDQ GHVLJQ )RU LQVWDQFH DQ XUEDQ GHVLJQ SODQ IRU D SDUWLFXODU DUHD
RI WKH FLW\ FDQ EH JHQHUDWHG E\ LPSRUWLQJ WKH DLPHG FLW\ PDS 8VLQJ WKH WUDLQHG
PRGHO E\ D PRUH VXEVWDQWLDO FLW\ WR SUHGLFW WKH EXLOGLQJV LQ D VPDOOHU FLW\ ZRXOG
JLYH XV HQFRXUDJLQJ UHVXOWV IRU WKH IXWXUH GHYHORSPHQW RI WKH VPDOOHU FLW\ 7KLV
PHWKRGRORJ\ LV HIILFLHQW DQG DFFXUDWH WR JHQHUDWH XUEDQ GHVLJQ SODQ VROXWLRQV
$OVR ZLWK WKH GHYHORSPHQW RI *$1 WKH PDFKLQH OHDUQLQJ DOJRULWKPV PD\
DVVLVW PRUH LQ WKH IXWXUH RI WKH XUEDQ GHVLJQ SURFHVV VLQFH *$1 ZLWK D UREXVW
GDWDEDVH WKDW FDQ EH XVHG WR FXVWRPL]H XUEDQ SODQQLQJ RI D FLW\ LQ WKH EHJLQQLQJ
VWDJH RI WKH XUEDQ GHVLJQ
%HVLGHV WKH JHQHUDWLYH DQDO\VLV EDVHG RQ WKH JHQHUDWHG UHVXOWV ZLWK GLIIHUHQW
PRGHOV DQG WKH VDPH VLWH FRQGLWLRQ SURYLGHV D PHWKRG WR HYDOXDWH DQ XUEDQ
GHVLJQ DQG FRQFOXGH WKH GHVLJQ VWUDWHJLHV XQGHU WKH VDPH FLUFXPVWDQFH ZKLFK
LV LPSRVVLEOH E\ WKH WUDGLWLRQDO DQDO\WLFDO PHWKRGV
5HIHUHQFHV
$JDUZDO 6 'DYH 5 DQG %DVVHWW %$  3DLQWLQJ JDOD[LHV LQWR GDUN PDWWHU KDORHV XVLQJ
PDFKLQH OHDUQLQJ 0RQWKO\ 1RWLFHV RI WKH 5R\DO $VWURQRPLFDO 6RFLHW\ 
&KHQJ &, DQG /LX '60  $Q LQWHOOLJHQW FORWKHV VHDUFK V\VWHP EDVHG RQ IDVKLRQ VW\OHV
 ,QWHUQDWLRQDO &RQIHUHQFH RQ 0DFKLQH /HDUQLQJ DQG &\EHUQHWLFV 
*RRGIHOORZ , 3RXJHW$EDGLH - 0LU]D 0 ;X % :DUGH)DUOH\ ' 2]DLU 6 &RXUYLOOH
$ DQG %HQJLR <  *HQHUDWLYH DGYHUVDULDO QHWV $GYDQFHV LQ QHXUDO LQIRUPDWLRQ
SURFHVVLQJ V\VWHPV 
+XDQJ : DQG =KHQJ +  $UFKLWHFWXUDO GUDZLQJV UHFRJQLWLRQ DQG JHQHUDWLRQ WKURXJK
PDFKLQH OHDUQLQJ 3URFHHGLQJV RI WKH WK $QQXDO &RQIHUHQFH RI WKH $VVRFLDWLRQ IRU
 - 6+(1 (7 $/
&RPSXWHU $LGHG 'HVLJQ LQ $UFKLWHFWXUH 0H[LFR &LW\ 0H[LFR 
.LP +6 DQG &KR 6%  $SSOLFDWLRQ RI LQWHUDFWLYH JHQHWLF DOJRULWKP WR IDVKLRQ GHVLJQ
(QJLQHHULQJ DSSOLFDWLRQV RI DUWLILFLDO LQWHOOLJHQFH 
.YRFKLFN 7  6QHDN\ 6SDWLDO 6HJPHQWDWLRQ 5HDGLQJ DUFKLWHFWXUDO GUDZLQJV ZLWK 'HHS
QHXUDO QHWZRUNV DQG ZLWKRXW ODEHOLQJ GDWD 3URFHHGLQJV RI WKH WK $QQXDO &RQIHUHQFH RI
WKH $VVRFLDWLRQ IRU &RPSXWHU $LGHG 'HVLJQ LQ $UFKLWHFWXUH 0H[LFR &LW\ 0H[LFR
0LUDEDO 1 &KDUOHV ( )HUUDUD ( *RQWKLHU 3 +DUGLQJ $ 6DQFKH]&RQGH 0$ DQG
7KRPSVRQ '  )*/ GHPRJUDSKLFV RXWVLGH WKH JDODFWLF SODQH XVLQJ VXSHUYLVHG
PDFKLQH OHDUQLQJ 3XOVDU DQG GDUN PDWWHU VXEKDOR LQWHUSUHWDWLRQV 7KH $VWURSK\VLFDO
-RXUQDO 
3DJDQLQL 0 GH 2OLYHLUD / DQG 1DFKPDQ %  &DOR*$1 6LPXODWLQJ ' KLJK HQHUJ\
SDUWLFOH VKRZHUV LQ PXOWLOD\HU HOHFWURPDJQHWLF FDORULPHWHUV ZLWK JHQHUDWLYH DGYHUVDULDO
QHWZRUNV 3K\VLFDO 5HYLHZ ' 
6WHLQIHOG . 3DUN . 0HQJHV $ DQG :DONHU 6  )UHVK (\HV $ IUDPHZRUN IRU WKH
DSSOLFDWLRQ RI PDFKLQH OHDUQLQJ WR JHQHUDWLYH DUFKLWHFWXUDO GHVLJQ DQG D UHSRUW RI DFWLYLWLHV
DW 6PDUWJHRPHWU\  3URFHHGLQJV RI &$$' )XWXUHV 
:DQJ 7& /LX 0< =KX -< 7DR $ .DXW] - DQG &DWDQ]DUR %  +LJKUHVROXWLRQ
LPDJH V\QWKHVLV DQG VHPDQWLF PDQLSXODWLRQ ZLWK FRQGLWLRQDO JDQV 3URFHHGLQJV RI WKH ,(((
FRQIHUHQFH RQ FRPSXWHU YLVLRQ DQG SDWWHUQ UHFRJQLWLRQ 
=KHQJ +  'UDZLQJ ZLWK %RWV +XPDQFRPSXWHU &ROODERUDWLYH 'UDZLQJ ([SHULPHQWV
3URFHHGLQJV RI WKH UG ,QWHUQDWLRQDO &RQIHUHQFH RQ &RPSXWHU$LGHG $UFKLWHFWXUDO 'HVLJQ
5HVHDUFK LQ $VLD %HLMLQJ &KLQD
=KHQJ + DQG +XDQJ :  8QGHUVWDQGLQJ DQG 9LVXDOL]LQJ *HQHUDWLYH $GYHUVDULDO
1HWZRUN LQ $UFKLWHFWXUDO 'UDZLQJV 3URFHHGLQJV RI WKH UG ,QWHUQDWLRQDO &RQIHUHQFH RQ
&RPSXWHU$LGHG $UFKLWHFWXUDO 'HVLJQ 5HVHDUFK LQ $VLD %HLMLQJ &KLQD
... Thanks to the explosive development of deep learning and internet-of-things, the handful of methodologies and ubiquitously available geo-social, urban and mobile data provide us with a new data-driven perspective to re-investigate urban planning. There are considerable existing works related to automated urban planning (Wang et al. 2021a;Shen et al. 2020;Ye, Du, and Ye 2021;Wang et al. 2021b). For example, motivated by the remarkable success of deep image generation, (Wang et al. 2020) proposes a land-use configuration generation framework, namely LUCGAN, which can generate a land-use configuration automatically for an empty geographical area based on surrounding contexts. ...
... For instance, (Khansari, Mostashari, and Mansouri 2014) studied the im-pact of the smart city on urban sustainability and urban planning. Recently, the remarkable success of deep learning has led researchers to think about how to utilize artificial intelligence to improve the efficiency of urban planning (Shen et al. 2020). For example, (Shen et al. 2020) utilized a GAN model to fill the urban elements in road map figures to produce the final urban plan. ...
... Recently, the remarkable success of deep learning has led researchers to think about how to utilize artificial intelligence to improve the efficiency of urban planning (Shen et al. 2020). For example, (Shen et al. 2020) utilized a GAN model to fill the urban elements in road map figures to produce the final urban plan. Compared with these works, IH-Planner is more advanced automatically and practically. ...
Article
The essential task of urban planning is to generate the optimal land-use configuration of a target area. However, traditional urban planning is time-consuming and labor-intensive. Deep generative learning gives us hope that we can automate this planning process and come up with the ideal urban plans. While remarkable achievements have been obtained, they have exhibited limitations in lacking awareness of: 1) the hierarchical dependencies between functional zones and spatial grids; 2) the peer dependencies among functional zones; and 3) human regulations to ensure the usability of generated configurations. To address these limitations, we develop a novel human-instructed deep hierarchical generative model. We rethink the urban planning generative task from a unique functionality perspective, where we summarize planning requirements into different functionality projections for better urban plan generation. To this end, we develop a three-stage generation process from a target area to zones to grids. The first stage is to label the grids of a target area with latent functionalities to discover functional zones. The second stage is to perceive the planning requirements to form urban functionality projections. We propose a novel module: functionalizer to project the embedding of human instructions and geospatial contexts to the zone-level plan to obtain such projections. Each projection includes the information of land-use portfolios and the structural dependencies across spatial grids in terms of a specific urban function. The third stage is to leverage multi-attentions to model the zone-zone peer dependencies of the functionality projections to generate grid-level land-use configurations. Finally, we present extensive experiments to demonstrate the effectiveness of our framework.
... The traditional design process, mainly relying on expert knowledge and design experience is inherently of low efficiency. It is particularly time-consuming in the early design stages, where designers need to iteratively and manually adjust the position and shape of each building to provide different layout solutions for reference [3]. To address these limitations, building layout generation, which uses computer-aided generative methods to automatically explore a wide range of design alternatives satisfying various design requirements, has been widely studied in both academia and industry in recent years [4][5][6][7][8][9][10][11][12][13]. ...
... For example, Wu et al. [27] conducted a comprehensive review of how GANs (generative adversarial networks) are applied to tackle challenging tasks in the built environment and identified 26 unique application domains enabled by GAN including floorplan generation, street-view generation, satellite image enhancement, etc. In terms of urban planning and building design, a few pioneering studies have started to apply deep generative methods including GAN and VAE (variational autoencoder) for automated building layout generation since 2020 [3,[27][28][29][30][31][32][33]. In contrast to rule-based methods and evolutionary algorithms, deep generative methods are developed based on deep neural networks to learn hidden relationships and urban knowledge from real-world design cases and synthesize visually realistic and semantically reasonable building solutions. ...
Article
Generative adversarial network (GAN) Design scenario 3D visualization A B S T R A C T Building layout generation has entered a new era in recent years, leveraging state-of-the-art deep generative methods to learn morphological properties of exiting urban structures and synthesize building alternatives responsive to local context. However, most existing research generally follows an image-to-image translation idea, while overlooking the impact of site/design attributes on building configuration, making their results less performative. Besides, most synthesized layouts are commonly displayed in 2D pixelized images, limiting further performance evaluation and informed decision-making. This study, therefore, proposes a novel GAN-based model, namely site-embedded generative adversarial networks (ESGAN) for automated building layout generation. Both qualitative and quantitative results in New York City indicate ESGAN is capable of synthesizing visually realistic and semantically reasonable layouts. This end-to-end generative system can not only encode a conditional vector to improve performance in different design scenarios but also display synthesized layouts at different levels of detail for human-system interaction.
... Compared with many other deep learning models, GANs can have extraordinary performances in high-resolution image classification and generation (Arjovsky, Chintala, & Bottou, 2017;Gonog & Zhou, 2019;Goodfellow et al., 2020;Radford, Metz, & Chintala, 2016), leading to a high potential of being applied to imageto-image translation tasks in architectural and planning research. For example, Shen, Liu, Ren, and Zheng (2020) used GANs to develop an algorithm of auto-filling building configurations based on the site conditions in city plans with high accuracy. They found GANs can be useful in assisting the design of urban forms in small cities and the accuracy may be promoted by building more robust databases. ...
Article
Land Surface Temperature (LST) is an important indicator of urban heat environments and can be largely influenced by the morphology factors of cities. However, previous studies mainly focus on large-scale and coarse-grained forecast modeling, making it hard to inform architects and urban designers without the advantage of quick, fine-grained prediction and visualization. The paper uses Generative Adversarial Networks (GAN) to address this gap by proposing a prediction model of city plans and corresponding LST heat maps. Taking New York City as an example, we use the Light Detection and Ranging (LiDAR) data, Landsat Surface Temperature data, and other relevant data to build seven hundred image pairs as the training set to train the model of predicting LST distribution. Using untrained pairs as the test set, the model can generate LST maps relatively quickly and accurately with the input of city plans. Then after accuracy analysis, different scenarios are simulated to test the application of the model in predicting the environmental impacts of plan modifications on land surface temperature. Eventually, the principles proposed in this paper can be applied to the development of relevant interactive design and planning tools in the future.
... Performance based (Rahimian, 2022), (Han, 2022), (Xu, 2022), (Jia, 2021), (Singh & Geyer, 2022), (Paterson, et al., 2017), (Baghdadi, et al., 2020), , (Li, et al., 2018), , (Singaravel, et al., 2018), , (Chokwitthaya, et al., 2019), (Gan, et al., 2019), (Olu-Ajayi, et al., 2022), (Zou, et al., 2021), (Li, et al., 2019), (Chou, Bui, 2014), (Wortmann, 2019), (Schwartz, et al., 2021), (Scherz, et al., 2022), (Sun, et al., 2015), (Mangan, 2021), (Ruiz, et al., 2017), (Singaravel, et al., 2018), (Toniolo, Leon, 2017), Multi objective optimization (Singaravel, et al, 2018), (Chardon, et al., 2016), (Natephra, et al., 2018), (Yousif & Bolojan, 2021), (Chardon, et al., 2015), (Liu, 2022), (Zhuang, et al., 2021), (Zhang, et al., 2021), (Baydoğan & Şener, 2014), (Chen & Pan, 2015), (Chen & Yang, 2017), (Si, et al., 2019), (Razmi, et al., 2022), (Carbonari, et al., 2019), (Kim & Clayton, 2020), (Yi, 2019), (Pilechiha, et al., 2020), (Mukkavaara & Shadram, 2021), (Marcolino, et al., 2015), (Seghier, et al., 2022) Spatial programming (Nisztuk & Myszkowski, 2019), (Ng, et al., 2019), (Buruzs, et al., 2022), (Karadoğan, 2021), (Doukari & Greenwood, 2020), (Raman & D'Souza, 2019), , , (Xiong, et al., 2022), (Liu & Lee, 2022), , (Shen, et al., 2020), (Yong & Chibiao, 2022), (Guo & Li, 2017), (Bei, et al., 2019), (Uzun, 2020), (Güleç, 2014), (Şen, 2022) Form finding , (Guo, 2022), (Cai & Li, 2021), (Radziszewski, 2017), , (Zheng & Yuan, 2021), , (Müezzinoğlu, 2022), (Aldemir, 2014), (Zheng, 2022) Restoration works (Gade, et al., 2018), (Mulero-Palencia, et al., 2021), (Morbidoni, et al., 2020), (Kamari, et al., 2018), (Jiang, et al., et al., 2022) Design tool development (Jalaei, et al., 2015), (Cichocka,et al., 2017), (Gade, et al., 2018) According to the algorithmic methodologies employed in their research, 242 studies have been categorized. The results of these studies suggest that genetic algorithm-based techniques are extensively employed in the published works ( Figure 6). ...
Article
Full-text available
Since the advent and usage of artificial intelligence approaches in architecture, a significant number of studies have focused on integrating technological solutions to architectural issues. Artificial intelligence applications in architectural design range from intelligent material design to architectural plan solutions. The ubiquity and distribution of research in this field, as well as the rising use of artificial intelligence techniques to solve design challenges, require an analytical classification of the essential literature review. This article presents a descriptive and analytical review of the work on artificial intelligence applications in architecture. A strong review has been made that identifies and addresses the gaps in artificial intelligence and architecture; and the literature review is transformed into statistical plots. The study's findings indicate a growing interest in artificial intelligence in the field of architecture. There is a need for novel research to be conducted in these areas using advanced technology and techniques.
... Huang W et al. used the cGAN model to effectively control the results to generate satellite images, hand-drawn architectural sketches, and architectural plan function zoning maps based on the boundary and functional training [32]. Shen et al. extracted urban information and simplified it into a floor plan and used cGAN to generate architectural layouts based on site conditions to achieve a rapid urban design method [33]. Cho Dahngyu et al. use cGAN to classify and recognize basic objects of floor plans. ...
Article
Full-text available
Existing methods for generating 2D plans based on intelligent systems usually require human-defined rules, and their operations are complex. GANs can solve these problems through independent research and learning. However, they only have generative design research based on a single constraint condition, and whether they can generate a qualified design scheme under many constraints is still unclear. Therefore, this paper develops the M-StruGAN generative model based on the structural design framework of a GAN. Its application research is extended to the 2D-plan layout generation of homestay based on the constraints of hybrid structures, and the feasibility of the method is comprehensively verified through three aspects: image synthesis quality assessment, scheme rationality assessment, and scheme design quality assessment. Experimental results show that the quality of the drawings generated by M-StruGAN is qualified, designers have a high degree of acceptance of the design results of M-StruGAN, and M-StruGAN completed the learning of the critical points of the 2D layout. Finally, through the human–computer interaction application of M-StruGAN, it can be found that compared with traditional design methods, M-StruGAN based on pix2pixHD has high-definition image quality, higher design efficiency, lower design cost, and more stable design quality.
... To obtain a high-definition map reflecting the built environment in Shanghai, we collect information on buildings, roads, green space, and waters and then we label them in different colors from the open platform API of BDmap by python (Shen et al., 2020). Labels function as supervisory signals that guide the network in processing, extracting, and transforming visual information to achieve maximum performance on the required tasks ( Figure 1). ...
Conference Paper
Full-text available
The working population has increased in cities with urbanization. Providing a supportive built environment with reasonable amenities distribution for them is becoming more important. Previous GIS-based approaches to urban planning for this issue tend to be subjective with high labour costs. This paper uses the generative adversarial network (GAN) to explore the relationship between amenities distributions and urban morphology, thus effectively predicting and visualizing the ideal amenities distributions in fast-growing cities based on the condition of well-developed megacities. In this research, we take Shanghai, one of the global cities in China with a big labour market, as the research sample. First, we use the Point of Interest (POI) data to draw the heatmap of urban amenities that support workers' daily life and collect the corresponding city maps. Then, we cut them into hundreds of image pairs as the training set and train a GAN model for predicting the future amenities distributions in other cities. To implement the model, we further collect the city maps of Jiaxing, one of the second-tier cities near Shanghai, as the testing set. Results show that our trained model can accurately predict amenities distributions for its future. The GAN-based prediction could effectively support detailed urban planning.
Presentation
Full-text available
利用条件生成对抗网络(CGAN) 构建一个通过城市形态及业态预测新冠疫情热点分布的方法,建立城市形态要素与新冠疫情之间的关系,使得机器可以自动预测城市中的疫情热点区域。以澳门为例,利用该方法确定了澳门城市形态及各类业态热点与新冠疫情足迹热点的相关性。
Article
Integrating deep generative models into urban form generation is an innovative and promising approach to support the urban design process. However, most deep generative urban form models are based on image representations that do not explicitly consider topological relationships among urban form elements. Toward developing an urban form generation framework aided by deep generative models and considering topological information, this paper reviews urban form generation, deep generative models/deep graph generation, and the state of the art of deep generative models in architectural and urban form generation. Based on the literature review, a topology-based urban form generation framework aided by deep generative models is proposed. The hypotheses of street network generation by deep generative models for graph generation and plot/building configuration generation by deep generative models/space syntax and the feasibility of the proposed framework require validation in future research.
Conference Paper
Full-text available
Machine learning has been proven to be a very efficient tool in urban analysis, using models trained with big data. We have seen research that applies a generative adversarial network (GAN) to train models, feeding the street map and visualized urban characteristics to predict certain urban features. However, in most cases, the input map is a two-dimensional (2D) map that only stores the land type data (e.g., building, street, green space), hence reducing building information to only the ground-floor area. The identities of buildings with similar floor areas can be hugely different, which may contribute to the prediction errors in previous machine-learning models. In this research, we emphasize the importance of the use of an image-based neural network to analyze the relationship between urban features and the constructed environment. We compare the model that uses traditional street color maps as the input set, against a new input set with more detailed building data. Once trained, the model with the enhanced input set yields output at a higher level of accuracy in certain areas. We apply the new model framework to three selected urban features predictions: rental price, building energy cost, and food sanitary ratio. A broad range of new research could be conduct with our new framework.
Conference Paper
Full-text available
With the development of information technology, the ideas of programming and mass calculating were introduced into the design field, resulting in the upcoming of computer-aided design. With the idea of designing by data, we began to manipulate data directly, and interpret data through design works. Machine Learning, as a decision making tool, has been widely used in many fields. It can be used to analyze large amount of data, and predict the future changes. Generative Adversarial Network (GAN) is a model frame in machine learning. It’s specially designed to learn and generate output data with similar or identical characteristics. Pix2pixHD is a modified version of GAN that learns image data in pairs and generate new images based on the input. The author applied pix2pixHD in recognizing and generating architectural drawings, marking rooms with different colors and then generating apartment plans through two convolutional neural networks. Next, in order to understand how these networks work, the author analyzed the frame of them, and provide an explanation of the three working principles of the networks, convolution layer, residual network layer and deconvolution layer. Lastly, in order to visualize the networks in architectural drawings, the author derived data from different layer and different training epochs, and visualized as gray scale images. It was found that the features of the architectural plan drawings have been gradually learned and stored as parameters in the networks. As the networks get deeper and the training epoch increases, the features in the graph become more concise and clearer. This phenomenon may be inspiring in understanding the designing behavior of human.
Article
We use machine learning (ML) to populate large dark matter-only simulations with baryonic galaxies. Our ML framework takes input halo properties including halo mass, environment, spin, and recent growth history, and outputs central galaxy and overall halo baryonic properties including stellar mass, star formation rate (SFR), metallicity, and neutral hydrogen mass. We apply this to the MUFASA cosmological hydrodynamic simulation, and show that it recovers the mean trends of output quantities with halo mass highly accurately, including following the sharp drop in SFR and gas in quenched massive galaxies. However, the scatter around the mean relations is under-predicted. Examining galaxies individually, at $z=0$ the stellar mass and metallicity are accurately recovered ($\sigma\lesssim 0.2$~dex), but SFR and HI show larger scatter ($\sigma\gtrsim 0.3$~dex); these values improve somewhat at $z=1,2$. Remarkably, ML quantitatively recovers second parameter trends in galaxy properties, e.g. that galaxies with higher gas content and lower metallicity have higher SFR at a given $M_*$. Testing various ML algorithms, we find that none performs significantly better than the others. Ensembling the algorithms does not fare better, likely because of correlations between the algorithms and the fact that none of the algorithms predict the large observed scatter around the mean properties. For the random forest, we find that halo mass and nearby ($\sim 200$~kpc) environment are the most important predictive variables followed by growth history. We find that halo spin and $\sim$Mpc scale environment are not. Finally we study the impact of additionally inputting key baryonic properties $M_*$, SFR, and $Z$, as would be available e.g. from an equilibrium model, and show that particularly providing the SFR enables HI to be recovered substantially more accurately.
Article
We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to 1/2 everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.
Conference Paper
This work presents an intelligent clothes search system based on domain knowledge, targeted at creating a virtual assistant to search clothes matched to fashion and userpsila expectation using all what have already been in real closet. All what garment essentials and fashion knowledge are from visual images. Users can simply submit the desired image keywords, such as elegant, sporty, casual, and so on, and occasion type, such as formal meeting, outdoor dating, and so on, to the system. And then the fashion style recognition module is activated to search the desired clothes within the personal garment database. Category learning with supervised neural networking is applied to cluster garments into different impression groups. The input stimuli of the neural network are three sensations, warmness, loudness, and softness, which are transformed from the physical garment essentials like major color tone, print type, and fabric material. The system aims to provide such an intelligent user-centric services system functions as a personal fashion advisor.