STAR-NT: Spatiotemporal Acceleration of Real-Time Neural Transparency Rendering
2026-06-15 • Graphics
GraphicsMachine Learning
AI summaryⓘ
The authors developed a way to make rendering scenes with transparent layers faster, especially on devices like phones. They use a smart method to adjust how much detail they process based on how much the colors change in different screen areas. They also reuse information from previous frames when possible, instead of recalculating everything each time. These improvements help keep the picture quality high while reducing the work needed for each frame.
neural order-independent transparencyspatiotemporal accelerationquadtreescreen-space subdivisioncolor variancedepth-based reprojectionreal-time renderinggeometry passtemporal coherence
Authors
Grigoris Tsopouridis, Christos Georgiou-Mousses, Aris Panagiotidis, Andreas Vasilakis, David Corrigan, Tobias A. Franke, Aleksei Gorbonosov, Andrei Astapov, Ioannis Fudos
Abstract
Neural order-independent transparency delivers high-quality rendering of overlapping transparent surfaces, but its geometry passes and network input generation remain costly, particularly on mobile and legacy hardware. We present a spatiotemporal acceleration framework that exploits spatial and temporal coherence to reduce this overhead while preserving visual quality. Spatially, we use adaptive quadtree-based screen-space subdivision to scale geometry pass resolution according to local color variance. Temporally, selected frames reuse the previous transparency result through depth-based reprojection instead of full rendering. Together, these optimizations reduce rendering cost and integrate efficiently into existing real-time rendering pipelines.