Visionary-Laboratory/visionary

What it solves

Visionary addresses the difficulty of rendering complex 3D neural representations—such as Gaussian Splatting (3DGS), 4DGS, and neural avatars—directly in a web browser. It provides a unified, high-performance platform that bridges the gap between advanced 3D AI research models and accessible, real-time web deployment.

How it works

The platform is built on WebGPU for high-performance parallel rendering and utilizes ONNX Runtime for per-frame neural inference. It features a standardized "Gaussian Generator" contract, allowing developers to plug in various algorithms to generate or update 3D particles dynamically. It also supports hybrid rendering, enabling the seamless mixing of Gaussian point clouds with traditional 3D meshes by managing depth compositing.

Who it’s for

This project is for researchers and developers working on 3D computer vision, neural rendering, and world models who need a lightweight, "click-to-run" browser-based environment to visualize, compare, and deploy their 3DGS-family models without requiring heavy desktop software.

Highlights

• Native WebGPU Acceleration: Delivers high-performance rendering of millions of Gaussian particles directly in the browser.
• Hybrid Architecture: Automatically handles occlusion and depth mixing between neural point clouds and standard 3D meshes.
• Universal Asset Support: A single interface to load various formats including PLY, SPLAT, GLB, and ONNX-based dynamic models.
• Extensible Design: Provides a standardized contract for integrating custom algorithms and generative post-processing networks.