What question did this study set out to answer?

This research aims to create stylistic imagery in virtual reality that is both legible and comfortable for users.

March 14, 2026International Journal of Reasoning-based Intelligent Systems0 citationsOpen Access

Immersive virtual reality art generation via neural style transfer

WPWei PuUniversity of Electronic Science and Technology of China

Key Points

This research aims to create stylistic imagery in virtual reality that is both legible and comfortable for users.
Developed a real-time pipeline for VR art generation using a feed-forward multi-style engine.
Implemented temporal and binocular stabilisers to reduce flicker and perceptual rivalry.
Utilized GPU-resident textures and mixed precision to manage latency effectively.
Achieved 75 FPS while maintaining about 15 ms per frame on RTX-class devices.
Reduced temporal flicker by approximately 31%.
Stereo divergence decreased to 0.012, ensuring better visual coherence.

Abstract

Stylised imagery in VR must meet comfort budgets while remaining legible.To address this challenge, a real-time pipeline is presented.First, a feed-forward multi-style engine applies post-lighting stylisation while head-locked interface layers are excluded.Then, temporal and binocular stabilisers curb flicker and rivalry.Finally, a render contract with GPU-resident textures, mixed precision, and deterministic fallback keeps latency under control.Experiments on three scenes sustain 75 FPS, about 15 ms per frame, and motion to photon of 19-23 ms on an RTX-class device; a fast baseline reaches 52 FPS at 22 ms and an iterative one 6 FPS at 140 ms.Image fidelity remains high, temporal flicker drops by roughly 31%, and stereo divergence falls to 0.012 with ninety-fifth percentile frame time near 17 ms.

AIに質問

Bookmark

View Full Paper