VGGT
Transforms single or multi-view images into rich 3D reconstructions, optimized for Apple Silicon with Metal Performance Shaders.
소개
VGGT-MPS is an optimized implementation of Facebook Research's Visual Geometry Grounded Transformer model, specifically engineered for Apple Silicon (M1/M2/M3) Macs utilizing Metal Performance Shaders (MPS). This tool empowers users to generate detailed 3D scene reconstructions from various image inputs, including predicting depth maps, camera poses, and dense 3D point clouds. Key advancements like sparse attention enable memory-efficient, city-scale reconstructions, while a unified CLI and Gradio web interface provide flexible interaction. It also features integration with Claude Desktop via the Model Context Protocol, making it a powerful solution for advanced computer vision tasks on macOS.
주요 기능
- Multi-View 3D Reconstruction generating depth maps, point clouds, and camera poses
- MPS Acceleration for Apple Silicon (M1/M2/M3) GPUs
- 58 GitHub stars
- Model Context Protocol (MCP) server integration for Claude Desktop
- Gradio web interface for interactive 3D reconstruction
- Sparse Attention for O(n) memory scaling, enabling city-scale 3D reconstruction
사용 사례
- Generating 3D point clouds and scenes from multi-view images for visualization or further processing
- Integrating 3D vision capabilities into AI agents and workflows via MCP
- Performing advanced 3D reconstruction from image datasets on Apple Silicon hardware
