LogosIntelligenceLab/PVG

What it solves

PVG addresses the challenge of reconstructing and rendering dynamic urban environments from video data. Traditional methods often struggle to accurately represent moving objects or maintain high-quality visuals when synthesizing new camera angles in complex, real-world street scenes.

How it works

The project uses a technique called Periodic Vibration Gaussian. It builds upon Gaussian Splatting to model both static backgrounds and moving objects within urban scenes. By incorporating periodic motion patterns, the system can effectively separate dynamic elements from the static environment, allowing for high-fidelity reconstruction and real-time rendering of complex street sequences.

Who it’s for

This project is intended for researchers and developers working in computer vision, autonomous driving, and 3D scene reconstruction who need to process dynamic video data from datasets like Waymo or KITTI.

Highlights

• Enables high-quality novel view synthesis for dynamic urban scenes.
• Supports automatic separation of dynamic objects from static backgrounds.
• Optimized for real-time rendering performance.
• Compatible with standard autonomous driving datasets including Waymo and KITTI.