摩尔线程赢图形顶会3DGS挑战赛大奖 自研LiteGS全面开源

Core Insights - The article highlights the achievement of Moore Threads at the SIGGRAPH Asia 2025 conference, where the company won a silver medal in the 3D Gaussian Splatting Reconstruction Challenge with its proprietary technology LiteGS, showcasing its strength in next-generation graphics rendering technology [1][18]. Group 1: 3D Gaussian Splatting Technology - 3D Gaussian Splatting (3DGS) is a revolutionary 3D scene representation and rendering technology that offers a remarkable balance between image quality, efficiency, and resource usage, significantly outperforming traditional NeRF by enhancing rendering efficiency by hundreds to thousands of times while maintaining realistic rendering quality [3][5]. - The technology has shown strong adaptability and scalability in areas such as ray tracing, real-time rendering for VR/AR, and multi-modal fusion, making it a key focus for both academia and industry [5]. Group 2: Competition Details - The 3DGS Reconstruction Challenge required participants to complete high-quality 3DGS reconstruction within 60 seconds using real terminal video sequences and imperfect camera trajectories, with PSNR and reconstruction speed as evaluation metrics [6]. - The results of the challenge have been made publicly available, allowing for transparency and authority in the rankings [6]. Group 3: Moore Threads' Performance - Moore Threads' AI team, competing under the identifier "MT-AI," achieved an average PSNR of 27.58 and a reconstruction time of 34 seconds, placing them among the top three teams in the competition [7][8]. - The company’s performance reflects its leading capabilities in 3DGS algorithm construction and hardware-software optimization [7]. Group 4: LiteGS Development - Moore Threads has developed the LiteGS library, which optimizes the training process of 3DGS, achieving significant improvements in training efficiency and reconstruction quality through a full-stack optimization approach [9][10]. - LiteGS can achieve up to 10.8 times training acceleration while reducing parameter count by over 50%, demonstrating its practical engineering applicability [15]. Group 5: Open Source and Future Directions - LiteGS has been fully open-sourced on GitHub to promote collaborative development and continuous evolution in 3D reconstruction and rendering technologies [18]. - The company plans to discuss the implications of 3DGS and other graphics intelligence technologies at the upcoming MUSA Developer Conference, emphasizing their role in advancing embodied intelligence and other cutting-edge fields [18].