Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Let's talk about Fakes

Let's talk about Fakes

C1595f6a99fc51c0fb8e04b54863dbeb?s=128

Leszek Rybicki

April 08, 2021
Tweet

Transcript

  1. Let's talk about Fakes Let's talk about all the good

    things And the bad things that may be
  2. Making Faces • Face Synthesis
 from nothing to face •

    Facial Attributes 
 modify age, gender, add glasses change hairstyle or color • Face Swap
 use the body and background of target, face of source • Facial Expression Transfer
 transfer only the expression or modify expression to match an impersonator or an audio track (reenactment)
  3. None
  4. Face Synthesis https://research.nvidia.com/publication/2017-10_Progressive-Growing-of

  5. https://thispersondoesnotexist.com/

  6. Facial Attributes

  7. https://www.faceapp.com/

  8. https://www.faceapp.com/

  9. Face Off (1997)

  10. Nicolas Cage Everywhere

  11. Reenactment Cafe 80s, Hill Valley, USA, (October 21, 2015)

  12. Jordan Peele (2018)

  13. Face2Face (2016)

  14. RGB-D tracking

  15. None
  16. None
  17. FaceSwapGAN (2018)

  18. None
  19. None
  20. None
  21. None
  22. None
  23. None
  24. None
  25. None
  26. FaceShifter (Dec 2019)

  27. None
  28. None
  29. None
  30. How to Spot a Fake

  31. https://www.kaggle.com/c/deepfake-detection-challenge

  32. uneven skin tone artifacts in hair says strange things warped

    edges How to Spot a Fake obstruction glitches
  33. That was a real photo.

  34. uneven skin tone artifacts in hair says strange things warped

    edges How to Spot a Fake obstruction glitches
  35. https://people.csail.mit.edu/mrub/vidmag/

  36. Why this is important

  37. We are hardwired to react to faces.

  38. None
  39. Especially to moving faces. 
 Even more so, emotional.

  40. None
  41. We have a tendency to fall for and spread shocking

    images.
  42. None
  43. None
  44. We react to images that speak to our tribal instincts

    or align with our political world view
  45. None
  46. None
  47. No matter how we try, how many tools we create

    to detect fakes, we WILL BE AFFECTED...
  48. ...the question is: HOW?

  49. None