Core Insights - Moore Threads won the silver award at the 3D Gaussian Splatting Reconstruction Challenge during SIGGRAPH Asia 2025 in Hong Kong, showcasing its self-developed technology LiteGS with outstanding algorithm capabilities and hardware-software optimization [1] Group 1: Technology and Innovation - 3D Gaussian Splatting (3DGS) is a revolutionary 3D scene representation and rendering technology that excels in 3D reconstruction and real-time rendering, with potential foundational value in broader AI applications [1] - The training process of 3DGS typically takes tens of minutes to hours, which limits its widespread application due to performance constraints [1] - Moore Threads developed the LiteGS library, achieving full-link collaborative optimization from the underlying GPU system to mid-level data management and high-level algorithm design [1] Group 2: Performance Metrics - LiteGS significantly leads in training efficiency and reconstruction quality, setting a new performance benchmark in the field [2] - LiteGS can achieve up to 10.8 times training acceleration while reducing the parameter count by over 50% when reaching the same quality level as current optimal solutions [2] - With the same parameter count, LiteGS surpasses mainstream solutions by 0.2–0.4 dB in PSNR metrics and reduces training time by 3.8 to 7 times [2] - For lightweight models, LiteGS requires only about 10% of the original 3DGS training time and 20% of the parameter count to achieve equivalent quality, demonstrating exceptional engineering practicality and technological foresight [2] - LiteGS is fully open-sourced on the GitHub platform [2]