Recommended
More Related Content
What's hot
What's hot (20)
Similar to Accommodation-invariant Computational Near-eye Displays - SIGGRAPH 2017
Similar to Accommodation-invariant Computational Near-eye Displays - SIGGRAPH 2017 (20)
More from StanfordComputationalImaging
More from StanfordComputationalImaging (8)
Recently uploaded
Recently uploaded (20)
Accommodation-invariant Computational Near-eye Displays - SIGGRAPH 2017
- 1. Accommodation-invariant Computational Near-eye Displays Robert Konrad1, Nitish Padmanaban1, Keenan Molner1, Emily A. Cooper2, Gordon Weztstein1 1 Stanford University, 2 Dartmouth College http://www.computationalimaging.o
- 3. Accommodation and Retinal Blur
- 4. Blur Gradient Driven Accommodation
- 5. Blur Gradient Driven Accommodation
- 6. Blur Gradient Driven Accommodation
- 7. Blur Gradient Driven Accommodation
- 8. Blur Gradient Driven Accommodation
- 9. Blur Gradient Driven Accommodation
- 11. Q: can we drive accommodation with stereoscopic cues by optically removing the retinal blur cue?
- 17. Current VR Displays: Vergence & Accommodation Mismatch!
- 18. Existing Approaches Multiplane Rolland et al., Applied Optics 2000 Akeley et al., SIGGRAPH 2004 Focal Surfaces Matsuda et al., SIGGRAPH 2017 Light Field Huang et al., SIGGRAPH 2015 Lanman et al., SIGGRAPH Asia 2013 Holographic Maimone et al., SIGGRAPH 2017 Adaptive Focus Sugihara et al., SID 1998 Liu et al., ISMAR 2008 Koulieris et al.,SIGGRAPH 2017 Padmanaban et al., PNAS 2017
- 23. How do we remove the blur cue?
- 24. Aperture Controls Depth of Field Image courtesy of Concept One Studios
- 25. Aperture Controls Depth of Field Image courtesy of Concept One Studios
- 26. Aperture Controls Depth of Field Image courtesy of Concept One Studios
- 27. Maxwellian-type (pinhole) Near-eye Displays Point Light Source
- 28. Maxwellian-type (pinhole) Near-eye Displays Severely reduces eyebox! Spatial Light Modulator Point Light Source
- 29. Focal Sweep EDOF Cameras: Dowski & Cathey, App. Opt. 1995 Nagahara et al., ECCV 2008 Cossairt et al., SIGGRAPH 2010 60Hz
- 37. PSF Engineering Spatially InvariantDepth Invariant
- 42. ∗
- 43. ∗ =
- 47. Target Conventional Target Image Conventional Display @ 1D
- 48. Target Conventional Target Image Conventional Display @ 3D
- 49. Target Conventional AI AI @ 3D Conventional Display @ 3D
- 60. Point Spread Functions Conventional AI AI 2 Plane AI 3 Plane 4D (0.25m) 3D (0.25m) 2D (0.5m) 1D (0.5m) 0D (∞)
- 61. Point Spread Functions Conventional AI AI 2 Plane AI 3 Plane 4D (0.25m) 3D (0.25m) 2D (0.5m) 1D (0.5m) 0D (∞) 3D 1D
- 62. Point Spread Functions Conventional AI AI 2 Plane AI 3 Plane 4D (0.25m) 3D (0.25m) 2D (0.5m) 1D (0.5m) 0D (∞)
- 63. MTF of Prototype Spatial Frequency (cycles/degree)
- 64. MTF of Prototype Spatial Frequency (cycles/degree)
- 65. MTF of Prototype Spatial Frequency (cycles/degree)
- 66. MTF of Prototype Spatial Frequency (cycles/degree)
- 67. MTF of Prototype Spatial Frequency (cycles/degree)
- 68. Q: can we drive accommodation with stereoscopic cues by optically removing the retinal blur cue?
- 69. Follow the target with your eyes 4D (0.25m) 0.5D (2m) User Study #1 11 participants
- 70. Stimulus
- 71. User Study #1
- 72. User Study #1
- 73. Look at each target 0.5D (2m) 4D (0.25m) User Study #2 12 participants
- 79. Ideal Accommodation Response Individual User Response
- 80. Discussion
- 82. Image courtesy of karma.com
- 83. Future Work – Multifocal Lenses
- 84. Robert Konrad Computational Imaging Lab Stanford University stanford.edu/~rkkonrad computationalimaging.org
- 94. 3D 2.5D 2D 1.5D 1D 0D ConventionalAIAI2-planeAI3-plane
Editor's Notes
- Displays different from other displays presented today, in that they don't rely on the retinal blur cue to drive accommodation
- If we look at conventional vr display Creates virtual images some fixed distance away, which is what everyone is attempting to get away from, because it causes our eyes to focus only to that distance. This design is not viable for anyone trying to support accommodation
- If we were to look inside one of these displays and focus to 25cm, we’d see quite a blurry image But as we focus closer and closer to the plane of the virtual image, we see a crisper image appear until we focus to the plane of the virutal image and see a sharp image Like autofocus
- If we were to look inside one of these displays and focus to 25cm, we’d see quite a blurry image But as we focus closer and closer to the plane of the virtual image, we see a crisper image appear until we focus to the plane of the virutal image and see a sharp image
- If we were to look inside one of these displays and focus to 25cm, we’d see quite a blurry image But as we focus closer and closer to the plane of the virtual image, we see a crisper image appear until we focus to the plane of the virutal image and see a sharp image
- If we were to look inside one of these displays and focus to 25cm, we’d see quite a blurry image But as we focus closer and closer to the plane of the virtual image, we see a crisper image appear until we focus to the plane of the virutal image and see a sharp image
- If we were to look inside one of these displays and focus to 25cm, we’d see quite a blurry image But as we focus closer and closer to the plane of the virtual image, we see a crisper image appear until we focus to the plane of the virutal image and see a sharp image
- This is effectively how our auto-focusing mechanism works, by taking advantage of the gradient in the perceived blur
- What is a point spread function?
- However, today, we propose a display type that will have a consistent perceived blur regardless of focus state Meaning that we refocus, we end up seeing the exact same image... Which breaks our conventional method of refocusing Point spread function is the impulse response of a system to a point light source.
- Let’s take a step back and understand how our visual system works. When we look around the real world, our eyes perform two actions simultaneously: vergence and accommodation. Vergence refers to the relative rotation of our eyeballs in their sockets. If I hold my finger up like this and look at it, my eyes rotate inwards. Simple enough.
- Let’s take a step back and understand how our visual system works. When we look around the real world, our eyes perform two actions simultaneously: vergence and accommodation. Vergence refers to the relative rotation of our eyeballs in their sockets. If I hold my finger up like this and look at it, my eyes rotate inwards. Simple enough.
- Let’s take a step back and understand how our visual system works. When we look around the real world, our eyes perform two actions simultaneously: vergence and accommodation. Vergence refers to the relative rotation of our eyeballs in their sockets. If I hold my finger up like this and look at it, my eyes rotate inwards. Simple enough.
- Cross-coupling -> real world
- In real world the consistent cross coupling allows our vergence and accommodation to converge faster Both systems are always driven to the same distance But this is not the case in current VR displays
- Systems support vergence, displays correct stereoscopic images Virtual image plane is fixed due to glass optics Mismatch between the binocular disparity and retinal blur cues, and this cross-coupling is now in conflict Mismatch headache, eye strain, reduced visual clarity Therefore the goal in general here, is to support multiple accommodative planes or even better a continuous range of accommodation distances.
- Underlying theme: produce realistic retinal blur cues to drive accommodation Not what we are trying to achieve
- Optically disable
- And ask a more fundamental question. Given that the accommodation state of the two eyes are linked, can the accommodation switch between these two planes? If so, then we could get two planes of accommodation by simply switching out one of the lenses in your favorite headset, and changing the rendering pipeline slightly to account for the change in magnification. That would be great! And the answer is to the question is ... No... But before we dive into our results, let me show you how we came up with this idea of monovision. It is actually a common alternative to bifocal lenses when treating presbyopia.
- Think of your eye as a camera
- Because of the incredibly wide depth of field, objects at all depths look the same, which is the same as removing the retinal blur cue! Pinhole camera to pinhole display Don’t need need constrict the pupil, we can just open up a very small exit pupil of the system
- Because of the incredibly wide depth of field, objects at all depths look the same, which Need pupil tracking + steering to support this system
- Because of the incredibly wide depth of field, objects at all depths look the same, which Need pupil tracking + steering to support this system
- Make this slide have an animation where the the retinal blur cues are superimposed and added eventually created a depth-invariant blur cue 60 Hz
- All images from 3D diopters
- All images from 3D diopters
- All images from 3D diopters
- here’s out take on adaptive focus display hardware others try to build smaller and smaller displays, we probably built the world’s biggest VR display here
- A clever combinatino of time-modulated backlight intensity and displayed images may be a viable approach to optimizing image resolution
- Strobe backlight during sweep Creates multiple virtual images Not multifocal Accommodation driven to plane closer to vergence distance
- Till now I’ve shown you the conventional display mode, with only 1 virtual image plane, and The accommodation invariant mode where we perform the focal sweep. But now with the strobe we can implement a 2-plane AI mode
- But now with the strobe we can implement a 2-plane AI mode You can see that the blur
- Explain what MTF is Industry-standard slanted-edge method of capturing the MTF
- Explain what MTF is Industry-standard slanted-edge method of capturing the MTF
- Explain what MTF is Industry-standard slanted-edge method of capturing the MTF
- Explain what MTF is Industry-standard slanted-edge method of capturing the MTF
- Explain what MTF is Industry-standard slanted-edge method of capturing the MTF
- And ask a more fundamental question. Given that the accommodation state of the two eyes are linked, can the accommodation switch between these two planes? If so, then we could get two planes of accommodation by simply switching out one of the lenses in your favorite headset, and changing the rendering pipeline slightly to account for the change in magnification. That would be great! And the answer is to the question is ... No... But before we dive into our results, let me show you how we came up with this idea of monovision. It is actually a common alternative to bifocal lenses when treating presbyopia.
- For all modes, a 6.2cm Maltese cross oscillated between 0.5 and 4 D (mean 2.25 D, amplitude 1.75 D) at 0.125 Hz
- For all modes, a 6.2cm Maltese cross oscillated between 0.5 and 4 D (mean 2.25 D, amplitude 1.75 D) at 0.125 Hz
- For all modes, a 6.2cm Maltese cross oscillated between 0.5 and 4 D (mean 2.25 D, amplitude 1.75 D) at 0.125 Hz Indicate disparity driven accommodation via the removal of focus cues in a near-eye display can be achieved, although the resulting accommodative gain is not quite as high as with natural focus cues. However, there are many depth cues at play here and we are mainly interested in the effect of binocular disparity specifically on accommodation (removing all other cues) To do so, we performed a second study
- Static target at 9 discrete depths 2 second blank period 3 second stimulus
- Interesting because we don’t see the step response we’d expect to see for the 2-plane and 3-plane modes. But then again this is averaged data When we look at some individual data plots, we see that some people see very strong response to the AI condition, while others so none at all. It would be interesting to investigate why there is this much discrepancy between users.
- And ask a more fundamental question. Given that the accommodation state of the two eyes are linked, can the accommodation switch between these two planes? If so, then we could get two planes of accommodation by simply switching out one of the lenses in your favorite headset, and changing the rendering pipeline slightly to account for the change in magnification. That would be great! And the answer is to the question is ... No... But before we dive into our results, let me show you how we came up with this idea of monovision. It is actually a common alternative to bifocal lenses when treating presbyopia.
- I want to start with the stereoscope … However, the basic optics of these systems have remained largely unchanged since their conception over a century ago. Stereoscope, 1800s, similar to today Two views to get 3D perception
- However, today, we propose a display type that will have a consistent perceived blur regardless of focus state Meaning that we refocus, we end up seeing the exact same image... Which breaks our conventional method of refocusing Point spread function is the impulse response of a system to a point light source.
- Why this is a computational display? Point spread function engineering