Deep Convolutional Food Styling

What does a good photo of food look like?
Can a machine tell?
Can we fix up photos of food using deep learning?

Transcript

1. %FFQ
   $POWPMVUJPOBM
   'PPE
   4UZMJOH !MVOBSEPH
    %FFQ
   $POWPMVUJPOBM
   'PPE
   4UZMJOH %FFQ
   $POWPMVUJPOBM
   'PPE
   4UZMJOH
2. ࣗݾ঺հ w ϨγΣοΫ
   w HJUIVC
   
   
   
   
   
   
   !MVOBSEPH
   w ϙʔϥϯυਓ
   w ΫοΫύουͰݚڀ։ൃ෦
   w ػցֶशϦαʔνϟʔ
   w ೲ౾৯΂ΒΕΔ
3. recipes and photos by users
4. ྉཧࣸਅΛ
   ඒຯͦ͠͏ʹࡱΔ
5. ޫ https://snappygoat.com
6. https://snappygoat.com ৭
7. https://snappygoat.com ίϯτϥετ
8. https://snappygoat.com ߏਤ
9. https://snappygoat.com ޫ୔
10. https://snappygoat.com എܠ
11. https://snappygoat.com ಁ໌ੑ
12. https://snappygoat.com ཱମੑ
13. 
    
    ɻɻɻ
14. 5VSJOH
    5FTU
15. νϡʔϦϯάɾςετ
17. 5VSJOH
    5FTU G D
18. 5VSJOH
    5FTU 哋▶
19. 5VSJOH
    5FTU Generative Adversarial Learning Generator Discriminator ( %
21. %FFQ
    $POWPMVUJPOBM
    (FOFSBUJWF
    "EWFSTBSJBM
    /FUXPSLT
22. %PXOTBNQMJOH features or small resolution image convolutional layer or layers RGB image input output
23. 6QTBNQMJOH upsampling CNN layer or layers RGB image features or small resolution image input output
24. &ODPEFS%FDPEFS D E image in Decoder Encoder image out feature space
25. 'VMMZ$POOFDUFE
    $MBTTJpFS YES NO classes data or features also called “Dense” 
 or “Linear” layer
26. ( % νϡʔϦϯάɾςετษڧձ
27. Generator Discriminator G MPPLT
    MFHJU
    UPUBMMZ
    TIPQQFE D
28. G SFBM
    GBLF D D(G(noise)) ˠ real (FOFSBUPS
    USBJOJOH Discriminator acts as the teacher
29. G SFBM
    GBLF D SFBM
    GBLF D D(G(noise)) ˠ fake D(photo) ˠ real %JTDSJNJOBUPS
    USBJOJOH Generator provides negative examples
31. $ZDMF
    ("/
32. w QZUPSDI
    IUUQTKVOZBO[HJUIVCJP$ZDMF("/
    w UFOTPSqPX
    IUUQTHJUIVCDPNWBOIVZ[$ZDMF("/5FOTPS'MPX
33. https://www.youtube.com/watch?v=9reHvktowLY
34. https://github.com/tatsuyah/CycleGAN-Models
35. w DIBJOFS
    IUUQTHJUIVCDPN"JYJMFDIBJOFSDZDMFHBO
36. (FOFSBUPS transformation network https://arxiv.org/abs/1603.08155
37. TVCUMF
    WJWJE … TVCUMF
    WJWJE %JTDSJNJOBUPS fully convolutional judges patches 
 of the input image https://arxiv.org/abs/1603.08155
38. "EWFSTBSJBM
    -PTT X F G Y TVCUMF
    WJWJE TVCUMF
    WJWJE … TVCUMF
    WJWJE TVCUMF
    WJWJE … X(F(low_score)) ˠ classify as high_score Y(F(high_score)) ˠ classify as low_score
39. $ZDMF
    -PTT G F G(F(image))ˠ the same image F G F(G(image))ˠ the same image
40. 5SBJOJOH
41. subtle vivid
54. *O
    UIF
    XJME
68. 'BJM
78. 6O*OTUB
    'JMUFS
79. https://snappygoat.com
80. https://snappygoat.com
81. https://snappygoat.com
82. Ҏ্