saguar1

YANAI Lab.

電気通信大学 総合情報学科/大学院 総合情報学専攻 メディア情報学コース 柳井研究室
電気通信大学 > 情報工学科 > コンピュータ学講座 > 柳井研究室 > 研究紹介  

$B%/%i%&%I%=!<%7%s%0$K$h$k?);v2hA|G'<1%b%G%k$N<+F09=C[(B

$BBg_7(B $BfF8c(B

2013$BG/(B 2$B7n(B 7$BF|(B

$B>\$7$/$O!$(B ($BO@J8(BPDF) ($B%9%i%$%I(B PDF) $B$r;2>H$7$F2<$5$$!%(B

1 $B8&5fGX7J!&8&5fL\E*(B

$B6aG/!$%/%i%&%I%=!<%7%s%0$r5!3#3X=,$K1~MQ$9$k8&5f$,A}2C$7$F$$$k!%(B $B%/%i%&%I%=!<%7%s%0$H$O!$(BWeb$B>e$NITFCDjB??t$N?M4V$K;E;v$r0MMj$9$k$3$H$G$"$k!%(B $B5!3#3X=,$N8&5f$K$*$$$F$O!$3X=,MQ%G!<%?%;%C%H$K%"%N%F!<%7%g%s$rIU2C$9$k:n6H$r%/%i%&%I%=!<%7%s%0$9$k$3$H$,B?$$!%(B $B=>Mh$O8&5f $BK\8&5f$G$O!$%/%i%&%I%=!<%7%s%0$rMQ$$$?%k!<%W3X=,$GJ*BN8!=P%b%G%k$r<+F09=C[$9$k:]!$%"%N%F!<%7%g%s:n6H$r0MMj$9$k2hA|$NA*=P@oN,$r(B3$B$D9M0F$7$?!%(B $B$3$N@oN,$,<}=8$5$l$?%G!<%?%;%C%H$H9=C[$5$l$?%b%G%k$K3F@oN,$,5Z$\$91F6A$r9M;!$9$k!%(B

2 $B4XO"8&5f(B

Vijayanarasimhan$B$i$N8&5f(B[1]$B$G$O%/%i%&%I%=!<%7%s%0$rMQ$$$FJ*BN8!=P4o!J@~7A(BSVM$B!K$N3X=,$KI,MW$J%G!<%?$r<}=8$7$F$$$k!%(B $B$3$N/?t$N%G!<%?$G(BSVM$B$r9=C[$7$?8e!$@~7A(BSVM$B$ND6J?LL$K6a$$%G!<%?$+$i=g$K%"%N%F!<%7%g%s$rIU2C$7$F$$$k!%(B $B$3$N@oN,$O!$>/?t$N3X=,%G!<%?$G9=C[$7$?@~7A(BSVM$B$N=PNO7k2L$r2aBgI>2A$7$F$$$k!%(B $B8=:_$N@~7A(BSVM$B$ND6J?LL$K6a$$%5%s%W%k$N65;U?.9f$rF@$k$H$$$&@oN,$O!$(BSVM$B$K$H$C$F[#Kf$J@~7A(BSVM$B$N3X=,$,?J$s$@8e$K$OM-8z$G$"$k!%(B $B$7$+$7!$>/?t$N3X=,%G!<%?$7$+$b$?$J$$@~7A(BSVM$B$N>l9g!$$=$ND6J?LL$+$i1s$$%5%s%W%k$KBP$9$k=PNO7k2L$,?.Mj$G$-$J$$>l9g$b$"$k$?$a!$D6J?LL$+$i1s$$%5%s%W%k$N%"%N%F!<%7%g%s7k2L$bF@$kI,MW$,$"$k!%(B

3 $BDs0F

$BDs0F
  1. $B8!=P%b%G%k$N=i4|2=!J?^(B1$B!K(B
    1. $B%-!<%o!<%I8!:w$rMQ$$$F2hA|$r<}=8$9$k!%(B
      $B%-!<%o!<%I$K$O!$8!=PBP>]$N%+%F%4%jL>$r;XDj$9$k!%(B
    2. $B<}=8$7$?2hA|$K(Bbounding box$B$rIUM?$9$k:n6H$r%/%i%&%I%=!<%7%s%0$9$k!%(B
    3. $B%/%i%&%I%=!<%7%s%0$7$?;E;v$N:n6H7k2L$rMQ$$$FJ*BN8!=P%b%G%k$r9=C[$9$k!%(B
  2. $B%k!<%W3X=,!J?^(B2$B!K(B
    1. $B8!=P%b%G%k$N=i4|2=$N:]$HF1MM$N
    2. $B<}=8$7$?2hA|$NCf$+$i!$(Bbounding box$B$NIA2h:n6H$r0MMj$9$k2hA|$rA*=P$9$k!%(B
    3. $BA*$S=P$7$?2hA|$K(Bbounding box$B$rIUM?$9$k:n6H$r%/%i%&%I%=!<%7%s%0$9$k!%(B
    4. $B%/%i%&%I%=!<%7%s%0$7$?;E;v$N:n6H7k2L$rMQ$$$F8!=P%b%G%k$r99?7$9$k!%(B

$BK\8&5f$G$O!$J*BN8!=P%b%G%k$K(BDeformable Part Model$B!J(BDPM$B!K(B[2]$B$rMQ$$$k!%(B DPM$B$O!$(BHOG$BFCD'NL$rMQ$$$?@~7A(BSVM$B%Y!<%9$NJ*BN8!=P4o$G$"$k!%(B DPM$B$N3X=,$K$O!$(Bbounding box$BIU$-$N2hA|$,I,MW$G$"$k!%(B $BJ,N`;~$K$O!$2hA|$rF~NO$9$k$H!$J*BN$N(Bbounding box$B$N0LCV$H$=$NI>2ACM!J(BSVM$B6-3&LL$H$N5wN%!K$,=PNO$5$l$k!%(B

Bounding box$B$NIA2h:n6H$r0MMj$9$k2hA|$rA*=P$9$k@oN,$H$7$F0J2<$N(B3$B$D$r:NMQ$7!$3F@oN,$,J*BN8!=P%b%G%k$KM?$($k1F6A$r9M;!$9$k!%(B

  • $B%i%s%@%`$KA*=P!J(BRandom$B@oN,!K(B
    $B$3$N@oN,$O%Y!<%9%i%$%s$H$7$F$NLr3d$r2L$?$9!%(B
  • $BI>2ACM$,8=:_$N8!=P%b%G%k$NogCM$K6a$$=g$KA*=P!J(BNear$B@oN,!K(B
    $B$3$N@oN,$O(BVijayanarasimhan$B$i$N8&5f(B[1]$B$G:NMQ$5$l$F$$$k!%(B $B8=:_$N8!=P%b%G%k$K$*$$$F!$I>2ACM$,Hs>o$KNI$$$b$N$d0-$$$b$N$KBP$9$k=PNO7k2L$O?.Mj$G$-$k$N$G!$%b%G%k$K$H$C$F[#Kf$J%5%s%W%k$N65;U%G!<%?$rF@$k$Y$-$G$"$k$H$$$&2>Dj$K4p$E$$$F$$$k!%(B
  • $BI>2ACM$,8=:_$N8!=P%b%G%k$NogCM$+$i1s$$=g$KA*=P!J(BFar$B@oN,!K(B
    $B$3$N@oN,$O!$=i4|2=$5$l$?8!=P%b%G%k$N=PNO7k2L$O$"$^$j?.Mj$G$-$J$$$N$G!$I>2ACM$,%b%G%k$NogCM$+$i1s$$%5%s%W%k$N65;U%G!<%?$rF@$k$Y$-$G$"$k$H$$$&2>Dj$K4p$E$$$F$$$k!%(B $BI>2ACM$,ogCM$+$i1s$$2hA|$KBP$9$k=PNO7k2L$,4V0c$C$F$$$?>l9g!$$=$N>pJs$r8!=P%b%G%k$KH?1G$5$;$l$P!$8!=P%b%G%k$OBg$-$/JQ99$5$l$k!%(B

Figure 1: $B8!=P%b%G%k$N=i4|2=(B
1#1

Figure 2: $B%k!<%W3X=,(B
2#2

4 $B

$BDs0FO$G<($7$?(B3$B$D$N@oN,$r:NMQ$7!$3F%k!<%W$K$*$$$F!$%]%8%F%#%V3X=,2hA|$NKg?t$H!$%F%9%H2hA|$KBP$9$k(BDPM$B$N(BF$BCM!J$H$b$K(B5$B%+%F%4%j$NJ?6Q!K$r5a$a$?!%(B $BK\]$H$9$k(B5$B$D$N%+%F%4%j$rI=(B4$B$K<($9!%(B


Table 1: $B8!=P%b%G%k$r9=C[$7$?%+%F%4%j(B
$B5mP'(B $BFy$8$c$,(B $B$*9%$_>F$-(B $B%i!<%a%s(B $B$?$$>F$-(B

$B3F%k!<%W$K$*$1$k%]%8%F%#%V3X=,2hA|$NKg?t$r?^(B3$B$K<($9!%(B Random$B@oN,$,:G$bDc$/!$(BNear$B@oN,$O%k!<%W$r=E$M$k$HDc2<$7!$(BFar$B@oN,$O$=$N5U$N798~$K$"$j!$:G=*E*$K$O(BFar$B@oN,$,:G$bB?$/$N%]%8%F%#%V2hA|$r<}=8$7$?!%(B $B$3$l$O!$=i4|>uBV$N(BDPM$B$O3X=,2hA|$,>/$J$$$?$a!$I>2ACM$,ogCM$K6a$$%]%8%F%#%V2hA|$,B8:_$9$k$,!$3X=,$,?J$`$K$D$l$F%]%8%F%#%V2hA|$NB?$/$OI>2ACM$,9b$/$J$j!$ogCM$+$i1s$6$+$k$+$i$@$H9M$($i$l$k!%(B $B$f$($K!$(BNear$B@oN,$G$O!$I>2ACM$,ogCM$+$i1s$6$+$C$?%]%8%F%#%V2hA|$rA*=P$G$-$J$$0lJ}!$(BFar$B@oN,$G$OI>2ACM$,9b$$$b$N$HDc$$$b$N$rA*=P$9$k$?$a!$3X=,$,?J$s$G$bI>2ACM$,9b$/$J$C$?%]%8%F%#%V2hA|$rA*=P$G$-$k!%(B

Figure 3: $B3F%k!<%W$K$*$1$k%]%8%F%#%V3X=,2hA|$NKg?t(B
3#3

$B3F%k!<%W$K$*$1$k(BDPM$B$N(BF$BCM$r?^(B4$B$K<($9!%(B $B%k!<%W3X=,$K$*$$$F$O!$%k!<%W$r=E$M$k$H(BF$BCM$O8~>e$9$k$3$H$,4|BT$5$l$k$,!$$I$N@oN,$K$*$$$F$b!$(BF$BCM$O2#$P$$798~$K$"$k$3$H$,<($5$l$F$$$k!%(B $B$3$l$O!$%"%N%F!<%7%g%s%G!<%?$K4^$^$l$k%N%$%:$,860x$G!$%k!<%W$r=E$M$k$H!$:F8=N($O>e>:$9$k$,E,9gN($O8:>/$9$k$+$i$G$"$k!%(B $B$?$$>F$-$rNc$K5s$2$k$H!$?^(B5$B$N25$B$N>eCJ$K<($9$h$&$JDL>o$N$?$$>F$-2hA|$H$O6KC<$K

Figure 4: $B3F%k!<%W$K$*$1$k(BF$BCM(B
4#4

Figure 5: $BDL>o$N$?$$>F$-2hA|$HF$-2hA|(B
5#5

$B%"%N%F!<%7%g%s%G!<%?$K%N%$%:$,4^$^$l$k$N$O!$%"%N%F!<%7%g%s:n6H$r0MMj$7$?:n6He$,%$%s%I?M$G!$;D$j$O>/?t$N%"%a%j%+?M$d%U%#%j%T%s?M$J$I$G$"$j!$F|K\?M$O$[$H$s$I$$$J$$!%(B $B$=$N$?$a!$:n6HpJs$r<}=8$7$?8e!$:n6H$Ko$K9b$$%"%N%F!<%7%g%s@:EY$r4|BT$9$k$3$H$OFq$7$$!%(B

5 $B$*$o$j$K(B

$BK\8&5f$G$O!$%k!<%W3X=,$rMQ$$$?J*BN8!=P4o$N<+F0@8@.$N:]!$3X=,2hA|$KDI2C$9$k2hA|$rA*Br$9$k(B3$B$D$N@oN,!J(BRandom, Near, Far$B!K$rHf3S$7$?!%(B $B$=$N7k2L!$$I$N@oN,$G$b%b%G%k$N@-G=$KM-0U:9$O8+$i$l$J$+$C$?$,!$(BFar$B@oN,$rMQ$$$?>l9g!$B>$N@oN,$KHf$Y$F%]%8%F%#%V2hA|$,B?$/=8$^$k$3$H$r3NG'$7$?!%(B

$BK\O@J8$G9T$C$?e$7$F$$$/$3$H$,4|BT$5$l$F$$$?$,!$I,$:$7$b$=$&$J$i$J$$$3$H$r3NG'$7$?!%(B $B$3$l$O!$%"%N%F!<%7%g%s%G!<%?$K%N%$%:$,B?$/4^$^$l$k$3$H$,860x$@$H;W$o$l$k!%(B $B%"%N%F!<%7%g%s%G!<%?$K4^$^$l$k%N%$%:$r=|5n$9$k$?$a$N=>Mh3,4]$B$O!$B?CMJ,N`$J$I$N4JC1$J:n6H$rBP>]$K$7$F$*$j!$$3$l$i$r(Bbounding box$BIA2h:n6H$H$$$&J#;($J:n6H$KE,MQ$9$k$3$H$OFq$7$$$?$a!$?7$7$$%N%$%:=|5n Bibliography

1
S. Vijayanarasimhan and K. Grauman.
Large-scale live active learning: Training object detectors with crawled data and crowds.
In Proc. of IEEE Computer Vision and Pattern Recognition, pp. 1449-1456, 2011.

2
P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan.
Object detection with discriminatively trained part based models.
IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 32, No. 9, pp. 1627-1645, 2010.

3
V. C. Raykar and S. Yu.
Ranking annotators for crowdsourced labeling tasks.
In Advances in Neural Information Processing Systems, pp. 1809-1817, 2011.

4
P. Y. Hsueh, P. Melville, and V. Sindhwani.
Data quality from crowdsourcing: A study of annotation selection criteria.
In Proc. of NAACL HLT Workshop on Active Learning for Natural Language Processing, pp. 27-35, 2009.