Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

[CHI 2018] OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation

91 views

Published on

http://koyama.xyz/project/optimo/

Published in: Technology
  • Be the first to comment

  • Be the first to like this

[CHI 2018] OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation

  1. 1. OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation Yuki Koyama & Masataka Goto
  2. 2. Optimization Techniques: - Powerful - Used in computer science and engineering https://en.wikipedia.org/wiki/Mathematical_optimization
  3. 3. How can animators utilize optimization techniques for motion editing effectively?
  4. 4. We propose “Optimization-Guided Motion Editing”
  5. 5. Background: How Do Animators Create Motions?
  6. 6. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 How Do Animators Create Motions? 7 1. Keyframing 2. Curve Editing
  7. 7. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 How Do Animators Create Motions? 8 1. Keyframing 2. Curve Editing Define key poses at several keyframes Adjust interpolation curves 
 (how key poses are interpolated between keyframes) frame = 4 frame = 14 … Time
  8. 8. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 How Do Animators Create Motions? 9 1. Keyframing 2. Curve Editing Define key poses at several keyframes frame = 4 frame = 14 … Our Target Adjust interpolation curves 
 (how key poses are interpolated between keyframes)
  9. 9. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Curve Editing is Important for Nuances 10 frame = 4 frame = 14 frame = 24 frame = 34 Even with the same keyframing, motions could be very different in nuances
  10. 10. Difficulties in Curve Editing
  11. 11. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Difficulties in Curve Editing 12 Unintuitive (indirect) effects Need to play motions every time manipulating handles Many parameters High-dimensional search task
  12. 12. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Difficulties in Curve Editing 13 Unintuitive (indirect) effects Need to play motions every time manipulating handles Many parameters High-dimensional search task
  13. 13. A Possible (?) Solution: 
 Naïve Optimization
  14. 14. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Optimization for Full Automation? 15 Optimization can fully automate the task, 
 but results are not always satisfactory (no control…)
  15. 15. Optimization-Guided Motion Editing Live Demo!
  16. 16. Design Goals: What Should be Considered for Effective Interaction?
  17. 17. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 18 ? Blackbox Original motion Optimal motion
  18. 18. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 19 1. Transparent Original motion Optimal motion
  19. 19. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 20 1. Transparent Original motion Optimal motion Feature: Optimization Process Visualization (for Transparency)
  20. 20. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 21 1. Transparent Original motion Optimal motion Feature: Cost Visualization (for Transparency)
  21. 21. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 22 1. Transparent Original motion Optimal motion
  22. 22. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 23 1. Transparent 2. Controllable Original motion Optimal motion
  23. 23. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 24 1. Transparent 2. Controllable Original motion Optimal motion Feature: Interactive Regularization (for Controllability)
  24. 24. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 25 1. Transparent 2. Controllable Original motion Optimal motion Feature: Time-Varying Cost Control (for Controllability)
  25. 25. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 26 1. Transparent 2. Controllable Original motion Optimal motion
  26. 26. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 27 1. Transparent 2. Controllable … 3. Editable Manual revision allowed Original motion Optimal motion
  27. 27. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 28 1. Transparent 2. Controllable … 3. Editable Manual revision allowed Original motion Optimal motion Opt. Control handles Preserved Feature: Preservation of Control Handles (for Editability)
  28. 28. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 29 1. Transparent 2. Controllable … 3. Editable Manual revision allowed Original motion Optimal motion Opt. Control handles Preserved Different arrangement Different data structure Feature: Preservation of Control Handles (for Editability)
  29. 29. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Three Design Goals 30 1. Transparent 2. Controllable … 3. Editable Manual revision allowed Original motion Optimal motion
  30. 30. Math: How to Formulate Optimization with Controllability
  31. 31. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Optimization with Controllability 32 min x ⇢Z C(x, t)dtTypical optimization
  32. 32. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Optimization with Controllability 33 min x ⇢Z wcost (t)C(x, t)dt + wreg D(x, xinit ) min x ⇢Z C(x, t)dt Control by time-varying cost weight Control by regularization weight Typical optimization Controllable optimization
  33. 33. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Optimization with Controllability 34 min x ⇢Z wcost (t)C(x, t)dt + wreg D(x, xinit ) min x ⇢Z C(x, t)dt Control by time-varying cost weight Control by regularization weight Adjusted in real time Typical optimization Controllable optimization
  34. 34. Example Usage Scenarios
  35. 35. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Example: Fox Tail • #parameters: 27 • Rig: Forward kinematics (FK) 36
  36. 36. Example: Fox Tail Initial motion - Too robotic… Let’s apply the (naïve) optimization first
  37. 37. Example: Fox Tail Naïve optimization - Too much changed! Let’s control it by regularization
  38. 38. Example: Fox Tail Optimization with regularization - Looks good!! Inspiration: How about making the swing speedier?
  39. 39. Example: Fox Tail Optimization with regularization & time-varying cost Time-varying cost weight
  40. 40. Validation: Interview with Professional Animators
  41. 41. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Informal Interview Goals • Validation of our approach (design goals, interactions, etc.) Participants • 2 professional animators (A1 & A2) who are familiar with 3D character animation authoring Procedure • Explain our design goals, explain our system features, and ask feedback comments (approx. 1 hour) 42
  42. 42. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Results Editability • Both A1 and A2 loved editability, which was described “indispensable” (A1) Transparency • We observed that both A1 and A2 easily understood the concept of optimization by just seeing the visualizations Controllability • All features on controllability were strongly appreciated • The interactivity makes the adjustment “easier” and “less stressful” (A1) 43
  43. 43. Discussions
  44. 44. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Related Work: Motion Editing Techniques 45 [Kazi+16][Glauser+16] Tangibles [Choi+16] Sketches Principles Optimization Ours
  45. 45. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Related Work: Design Interface with Optimization 46 [Koyama+14] [Umetani+14] Ours Photo color adjustment, etc. Paper airplanes Character motions [O’Donovan+15] Graphic layout
  46. 46. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Future Work 47 E.g., architecture design [Kilian+17]E.g., Autodesk Maya Integration to & evaluation in professional workflow Applications to 
 other design domains
  47. 47. Summary
  48. 48. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Optimization-Guided Motion Editing • Optimization is used for enhancing manual editing • OptiMo allows animators to interactively control optimization 49 Animator’s control less ive ng Interactive control by animators Automatic adjustment by optimization
  49. 49. Yuki Koyama and Masataka Goto. OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation. CHI 2018. http://koyama.xyz/project/optimo/ Slides & videos available! We also plan to release our source codes!
  50. 50. OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation Yuki Koyama & Masataka Goto
  51. 51. Appendix
  52. 52. Related Work
  53. 53. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Motion Editing Techniques — Automatic 54 [Witkin+88] [Merel+17] Optimization with space-time constraints Use of learned motion controllers Difficult to reflect artistic intention ➡ Lack of controllability Resulting motions do not have curve-based data representation ➡ Lack of editability
  54. 54. OptiMo’s Features
  55. 55. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 We Designed 6 Features for OptiMo 56 1. Preservation of control handles [ T & E ] 2. Optimization process visualization [ T ] 3. Cost visualization [ T ] 4. Interactive regularization [ T & C ] 5. Per-Joint regularization [ C ] 6. Time-varying cost control [ C ] Editability Transparency Controllability
  56. 56. Algorithm: How Does Optimization Work?
  57. 57. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Typical Optimization Formulation 58 min x ⇢Z C(x, t)dt Cost function C(x, t) Search parameters x
  58. 58. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Parametrization (Search Space) 59 Time Rig value Use the control handle parameters (that animators otherwise manipulate manually) as the search parameters of the optimization ➡ This ensures editability (and also transparency)
  59. 59. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Cost Function: Our (Current) Choice • Is well-established and used in computer graphics and robotics (c.f. [Safonova+04]) • Is a summation of necessary torques (rotational forces) on each joint • Is useful for avoiding physically unfeasible motions 60 C(x, t) = X j2J k⌧j(x, t)k2
  60. 60. Validation: Interview with Professional Animators
  61. 61. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Comments on Editability • Both A1 and A2 loved editability • A1: Editability is “indispensable” • This comment especially validates our parametrization: • A2: It is acceptable for “the [tangent] handles to change,” but “the keyframes (key points) should not be change” by optimization 62
  62. 62. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Comments on Transparency • Both A1 and A2 loved the visualization features • A1: The process visualization “is very interesting” • We observed that both A1 and A2 easily understood the concept of optimization by seeing the visualizations 63
  63. 63. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Comments on Controllability • All features on controllability were strongly appreciated • A2: “I would be puzzled if I can’t use this” • The interactivity of control features was appreciated • A1: These features could make adjustment “easier” and “less stressful” 64
  64. 64. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Comments on Usages in Production • The current system is not perfect yet, but it could be useful… • A2: For “sub-characters’ motion,” or • A1: In “projects with only limited time” • Both A1 and A2 got very excited about the (production-level) quality of the fox tail animation • A2: “It is possible to provide it by hand, but it requires much tweaking to make it (the motion) look natural” 65
  65. 65. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Other Comments • It would be nice if motion styles could be handled in the cost function as well as physics • A2: “Each art project has a specific style of motions” and “it would be nice if it automates [the process of applying such styles]” 66
  66. 66. Example Usage Scenarios
  67. 67. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Example: Stick Figure Waving His Arm • #parameters: 78 • Rig: Forward kinematics (FK) 68
  68. 68. Example 1: Stick Figure Waving His Arm Initial Naïve opt. Controlled opt. Too robotic… The spine/elbow bend wrongly… Looks natural :)
  69. 69. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Example: Fox Tail • #parameters: 27 • Rig: Forward kinematics (FK) 70
  70. 70. Example: Fox Tail Inspiration: Let’s make the swing speedier Time-varying cost control
  71. 71. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Example: Music Conductor • #parameters: 54 • Rig: Forward kinematics (FK) and inverse kinematics (IK) 72 The next slide contains audio!
  72. 72. Example: Music Conductor
  73. 73. Example: Music Conductor
  74. 74. Example: Music Conductor
  75. 75. Example: Music Conductor
  76. 76. Discussions
  77. 77. Yuki Koyama and Masataka Goto (AIST, Japan) | OptiMo: Optimization-Guided Motion Editing for Keyframe Character Animation | CHI 2018 Discussion: “One is not Enough” 78 Muscle models [Wang+12] Motion styles [Grochow+04] Providing more options of cost functions as well as our physics-based cost function — One is not enough!

×