Impostor-based Rendering Acceleration for Virtual, Augmented, and Mixed Reality
summary
Proposes runtime-generated stereoscopic impostors that store depth and use a per-eye ray‑marched parallax correction plus dilated depth and hit‑point multi-sampling to preserve binocular cues. Regeneration is decoupled across frames and prioritized by a regeneration metric combining conservative parallax and texture/disocclusion terms. Implemented in VR on a 324‑model, ~83M‑vertex test scene (≈9000 frames) and achieves >90 Hz and 3–6× speedup over naive mesh rendering (2–3× vs LOD), SSIM 0.98–0.99 and higher tPSNR than mesh with 4×MSAA; not suited for constantly changing geometry.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:22Impostors - Problematic in XR
- 1:50Related Work - Impostors
- 2:31Stereoscopic Impostor
- 3:13Parallax Corrected Impostor
- 4:02Reducing Aliasing Artifacts
- 4:50Regeneration System: Motivation
- 5:12Regeneration System: Overview
- 6:09Regeneration System: Temporal Blending
- 6:36Regeneration System: Regeneration Metric
- 7:56Evaluation - Performance
- 8:29Evaluation - Image Quality
- 8:53Evaluation - Temporal Stability
- 9:23Conclusion & Future Work
related talks
09mHow to Throw Out 95% of Pixels in VR, Without Anyone Noticing!!
19mDynamic Stereoscopic 3D Parameter Adjustment for Enhanced Depth Discrimination
27mHoloportation: Virtual 3D Teleportation in Real-time
07mBringing back human interaction to a digital world
41mComputer Vision for Remote AR
50mAdvanced Mixed Reality Development and Best Practices
46mScience based Design Decision Making for Virtual Reality
19mVirtual Replicas for Remote Assistance in Virtual and Augmented Reality