An optimized AI-driven 3D reconstruction pipeline with adaptive LOD and neural rendering for interactive cultural heritage visualization
Sep 2026 · Computer Vision, Graphics and Artificial Intelligence (CVGAI 2026) · Vol 14357, pp. 143570O - 143570O-6 · 0 citations
· 15 references
Engineering
Abstract
This study proposes an optimized AI-driven 3D reconstruction pipeline featuring: (1) improved cGAN with multi-scale perceptual and edge-aware losses; (2) adaptive LOD algorithm for real-time WebGL rendering; (3) hybrid neural rendering with progressive streaming. Experiments on 12 cultural relics demonstrate 28.5% improvement in geometric accuracy (Chamfer Distance), 42.3% FPS increase, and 38.7% loading reduction versus Instant-NGP baselines. IPA validation with 318 users confirms significant satisfaction enhancement in interaction fluidity and immersion.