Terminus Group Chief Scientist Dr. Ling Shao Unveils SOGS: A Landmark Breakthrough in 3D Gaussian Rendering at CVPR 2025

Terminus Group, a global leader in spatial intelligence technologies, announced a groundbreaking advancement in 3D scene rendering through its latest research paper, "SOGS: Second-Order Anchor for Advanced 3D Gaussian Splatting", accepted at CVPR 2025.

Dr. Ling Shao and his co-authors propose a novel second-order anchor technique, SOGS, that incorporates covariance-based second-order statistics and correlation across feature dimensions to augment features within each anchor, compensating for the reduced feature size and improving rendering quality effectively. In addition, the work introduces a selective gradient loss to enhance the optimization of scene textures and scene geometries, leading to high-quality rendering with small anchor features. Extensive experiments over multiple widely adopted benchmarks show that SOGS achieves superior rendering quality in novel view synthesis with clearly reduced model size.

The proposed SOGS can render high-quality textures and geometries and reduce model size simultaneously. The illustrations in (a) and (b) show Rendered Images and Gradient Maps, which are produced by Scaffold-GS and SOGS, respectively, for the samples ‘Stump’ and ‘Room’ from Mip-NeRF360. The Gradient Maps are extracted with the Sobel operator, which highlights the texture and geometry of image scenes.

SOGS addresses core limitations in current 3D Gaussian Splatting (3D-GS) methods—most notably, the trade-off between model size and rendering quality, the loss of local texture and geometric detail, and the imbalance in optimization across spatial regions.

Expanding Possibilities in Spatial Intelligence

As a versatile rendering framework, SOGS will empower a wide range of real-world applications by combining high visual fidelity with lightweight performance:

 Virtual & Augmented Reality (VR/AR)

· Enables real-time, high-frame-rate rendering on head-mounted displays, reducing latency and motion sickness.

· Compact model size allows seamless integration of AR content on mobile devices like smartphones and tablets.

Autonomous Driving & Robotics

· Rapidly reconstructs 3D scenes of roads, objects, and obstacles from sparse visual inputs.

· Supports real-time decision-making with compact, high-resolution 3D representations fused from LiDAR and other sensor data.

 Film Production & Game Development

· Speeds up real-time scene preview and visual effects workflows, offering a memory-efficient alternative to NeRF.

· Preserves intricate scene details—such as foliage and architectural features—while minimizing resource consumption in large-scale environments.

 Urban Planning & Architectural Visualization

· Enables fast generation of accurate, city-scale 3D models for simulations and planning tools, including traffic flow analysis and infrastructure design.

SOGS was officially presented during Poster Session 4 at CVPR 2025 in Nashville, where it drew widespread attention from researchers and industry leaders. The paper has been recognized for its practical impact and scientific innovation, signaling Terminus Group’s growing influence in the global AI research community.

Learn More:

https://cvpr.thecvf.com/Conferences/2025/AcceptedPapers