π^4 is a unified foundation model for 4D scene understanding that can recover dense geometry, camera poses, and metric scale from a single image, an unordered set of images, or a video. The model achieves the best average rank in overall geometry evaluation across multiple benchmarks, while remaining competitive in single-view and multi-view 2D/3D point tracking and moving object segmentation.
Source:Hugging Face 社区论文精选 · Read Original
Leave a Reply