Extending Spacewarp for Transparency: From Algebraic Blending to Alpha Compositing

Abstract

Minimizing motion-to-photon (M2P) latency is a paramount challenge in Extended Reality (XR) systems. Reprojection techniques such as Spacewarp (SW) decouple rendering framerates from display refresh rates by synthesizing intermediate frames based on depth-informed 3D warps of the final framebuffer. However, conventional single-pass implementations fail on transparent surfaces as the depth buffer typically encodes only the opaque background. As a result, transparent foreground elements exhibit visual instability, including jitter, shaking, and apparent detachment from their intended spatial positions during head pose updates. In this paper, we propose a comprehensive Dual-Layer Spacewarp Architecture that resolves these artifacts by separating opaque and transparent content into distinct rendering and reprojection passes. The framework incorporates an Off-Center Projection Model to support asymmetric VR frustums, and remains compatible with standard Spacewarp techniques with motion-vector correction. We evaluate two architectural variants: Variant-A (Algebraic Blending) reconstructs transparent surfaces analytically using differential snapshots, while Variant-B (Alpha Compositing) employs explicitly rendered, independent transparent buffers. To facilitate industry adoption, we further propose a standardized OpenXR extension, XR_KHR_composition_layer_transparency, which enables applications to submit multi-layer buffers directly to the runtime. Experimental results demonstrate that both variants successfully leverage dual-depth reprojection to handle multi-layer scenes. Notably, Variant-B achieves near ground-truth visual fidelity even under large pose prediction errors where single-pass reprojection methods fail.