Bag-of-visual-words を用いた 2クラスの画像識別器の実装 + ハンズオン

Slide 1

Slide 1 text

#BHPGWJTVBMXPSETΛ༻͍ͨ  Ϋϥεͷը૾ࣝผثͷ࣮૷ϋϯζΦϯ ෬ݟྒྷฏ

Slide 2

Slide 2 text

2 https://github.com/sidestepism/pilaf http://sidestepism.com/package.zip http://github.com/sidestepism/pilaf

Slide 3

Slide 3 text

՝୊ w νϟʔϋϯͱϐϥϑͷ൑ผ͸೉͍͠ɽ 3

Slide 4

Slide 4 text

࣮ࡍਓؒ͸Ͳͷ͘Β͍ ࣝผͰ͖Δͷ͔ 4

Slide 5

Slide 5 text

ݹాࣝผثWTઢܗࣝผث 5

Slide 6

Slide 6 text

Slide 7

Slide 7 text

Slide 8

Slide 8 text

Slide 9

Slide 9 text

Slide 10

Slide 10 text

Slide 11

Slide 11 text

Slide 12

Slide 12 text

Slide 13

Slide 13 text

Slide 14

Slide 14 text

Slide 15

Slide 15 text

Slide 16

Slide 16 text

՝୊ w ը૾͸௒ߴ࣍ݩͷϕΫτϧ ࣍ݩ w ͨͩ͠Ұൠͷը૾ʹ͸༷ʑͳ੍໿͕͋Γɼ  ͜ͷ੍໿ʹదԠ͢ΔΑ͏ʹਓྨ͸ࢹ஌֮Λൃୡ͖ͤͯͨ͞ w ࢹ஌֮తʹॏཁͳ৘ใΛ࢒ͨ͠··࣍ݩΛѹॖ͠ͳ͚Ε͹ͳΒͳ͍  Ͱ͖Ε͹ʙ࣍ݩ͘Β͍ 16

Slide 17

Slide 17 text

ہॴಛ௃ྔղੳ w ը૾ͷہॴಛ௃ྔΛղੳ w 4*'5 463' )BBSMJLF )0( w ಛʹճసɾ֦େॖখʹର͢Δෆมੑ͕ॏཁ w ྫ͑͹4*'5͸ Y Z ࠲ඪ࣍ݩͷ࣮਺ϕΫτϧ w 463'ಛ௃ྔநग़ˠը૾͕࣍ݩϕΫτϧͷଋʹͳΔ 17

Slide 18

Slide 18 text

νϟʔϋϯɾϐϥϑσʔληοτͷߏங w #JOH4FBSDIը૾ݕࡧ͔Βʮνϟʔϋϯʯʮϐϥϑʯͷը૾Λຕͣ ͭऔಘ w લॲཧ w YʹϦαΠζˡඞཁ͔ ֦େॖখීวͰ͸ w άϨʔεέʔϧԽ w ώετάϥϜฏۉԽˡর໌ʹ͍ͭͯෆมͰ͸ 18

Slide 19

Slide 19 text

Slide 20

Slide 20 text

ઢܗࣝผ·ͰͷಓͷΓ w େྔͷը૾Λऩू w ೖྗը૾Λ࣍ݩϕΫτϧ܈ʹม׵ w ಛ௃ྔۭؒͰϕΫτϧ܈ΛϕΫτϧྔࢠԽ LNFBOT (.. Λ༻͍ͯ ෼ྨ WJTVBMXPSET ͠ɼը૾શମͷώετάϥϜΛ࡞੒ w ͜ͷώετάϥϜΛը૾Λ୅ද͢ΔϕΫτϧͱͯ͠Έͳ͠ઢܗࣝผ 20 Jun Yang, Yu-Gang Jiang, Alexander G. Hauptmann, and Chong-Wah Ngo. 2007. Evaluating bag-of-visual-words representations in scene classiﬁcation.

Slide 21

Slide 21 text

େྔͷը૾ऩू w #JOH4FBSDI"1*Λ༻͍ͨ ແྉͰϦΫΤετ݄ w OPEFKTͰฒྻμ΢ϯϩʔυ w ໌Β͔ʹϐϥϑνϟʔϋϯͰͳ͍΋ͷΛਓྗͰআڈ w νϟʔϋϯ ϐϥϑ ೳ೥ྰಸ 21

Slide 22

Slide 22 text

େྔͷը૾ऩू w ը૾αΠζ͕େ͖͍΋ͷ΋͋ΔͷͰɼ*NBHFNBHJDLͷSFTJ[F@UP@pMMΛ࢖ͬ ͯYʹαΠζΛਖ਼نԽ w ΞεϖΫτൺ͸ҡ࣋ɼDSPQ

Slide 23

Slide 23 text

463'ಛ௃ྔΛܭࢉ w 0QFO$7ͷ࣮૷Λར༻ɼIFTTJBOͷᮢ஋Λʹઃఆˡ͜ͷࠜڌ ੑ ࣭ w ͭͷը૾͋ͨΓ͘Β͍ͷಛ௃఺ ࣍ݩϕΫτϧ ͕ݕग़ 23

Slide 24

Slide 24 text

ϕΫτϧྔࢠԽ w શը૾ͷಛ௃ྔϕΫτϧͷू߹ΛLNFBOTΞϧΰϦζϜͰΫϥελԽ L w த৺఺ͷΈΛอଘ DMVTUFSEBU w ϗϯτ͸(..ͱ͔Ͱڞ෼ࢄߦྻΛ࢒͓ͯ͘͠ͱΑΓਖ਼֬ʹͳΔ w (..ಉ࢜ͷڑ཭ܭࢉ͸ΧϧόοΫϥΠϒϥʔ৘ใྔͳͲ 24 http://tech.nitoyon.com/ja/blog/2009/04/09/kmeans-visualise/

Slide 25

Slide 25 text

͍͖ͳΓνϟʔϋϯϐϥϑ͸ՙ͕ॏ͍ ͦ΋ͦ΋ͪΌΜͱಈ͍ͯΔ͔Ͳ͏͔ࣗ৴ͳ͍ 25

Slide 26

Slide 26 text

؆୯ͳྫ͸ͳ͍͔ 26

Slide 27

Slide 27 text

νϟʔϋϯWTೳ೥ྰಸࣝผث 27

Slide 28

Slide 28 text

݁Ռ νϟʔϋϯೳ೥ྰಸ w $IBODF3BUF w 1FSDFQUSPO w -PHJTUJD3FHSFTTJPO w -JOFBS,FSOFM47. w 3#',FSOFM47. HBNNB w 3#',FSOFM47. HBNNB w 4JHNPJE,FSOFM47. w QPMZOPNJBM,FSOFM47. EFHSFF w QPMZOPNJBM,FSOFM47. EFHSFF w QPMZOPNJBM,FSOFM47. EFHSFF w ෬ݟ 28

Slide 29

Slide 29 text

νϟʔϋϯWTϐϥϑࣝผث 29

Slide 30

Slide 30 text

݁Ռ νϟʔϋϯϐϥϑ w $IBODF3BUF w 1FSDFQUSPO w -PHJTUJD3FHSFTTJPO w -JOFBS,FSOFM47. w 3#',FSOFM47. HBNNB w 3#',FSOFM47. HBNNB w 4JHNPJE,FSOFM47. w QPMZOPNJBM,FSOFM47. EFHSFF w QPMZOPNJBM,FSOFM47. EFHSFF w QPMZOPNJBM,FSOFM47. EFHSFF w ෬ݟ 30

Slide 31

Slide 31 text

3FRVJSFNFOUT w ը૾ऩूεΫϦϓτOPEFKT +BWB4DSJQU w OQNSFRVFTU BTZOD w AOQNJOTUBMMAͰΠϯετʔϧͰ͖·͢ w ෬ݟͷ#JOH4FBSDI"1*τʔΫϯΛຒΊࠐΜͰΔ w ը૾Ճ޻εΫϦϓτSVCZ w HFNSNBHJDL w ACVOEMFJOTUBMMAͰΠϯετʔϧ ཁ*NBHFNBHJDL w 463'ಛ௃ྔநग़$ w ཁ0QFO$7 9DPEFͰϏϧυ w ֶशεΫϦϓτ1ZUIPO w QBDLBHFOVNQZ TDJQZ TDJLJUMFBSO 31

Slide 32

Slide 32 text

32 ͓·͚ઢܗࣝผثʹνϟʔϋϯͱͯ͠  ೝࣝ͞Εͨೳ೥ྰಸͷը૾ ˠը૾ͷલॲཧ΍औΓআ͖Ͱ  ΋ͬͱਖ਼֬ʹॲཧͰ͖ΔՄೳੑ͕ࣔࠦ