T3lescope: Arbitrary-Resolution High-Fidelity Generative Surface Reconstruction from Images
T3lescope uses one shared generator in a coarse-to-fine cascade to reconstruct detailed 3D meshes across scene scales.
The paper says the method works from posed multi-view images without per-scene optimization. A coarse pass sets the layout, then finer spatial cells refine inherited geometry to recover surface detail. The authors report gains over feed-forward and generative baselines, and results matching or beating per-scene optimization on indoor, outdoor, and city-scale scenes. They also claim it handles sparse views, glossy surfaces, and transparent surfaces better than prior fixed-resolution generative approaches. HF Daily Papers' note
The paper says the method works from posed multi-view images without per-scene optimization. A coarse pass sets the layout, then finer spatial cells refine inherited geometry to recover surface detail. The authors report gains over feed-forward and generative baselines, and results matching or beating per-scene optimization on indoor, outdoor, and city-scale scenes. They also claim it handles sparse views, glossy surfaces, and transparent surfaces better than prior fixed-resolution generative approaches. HF Daily Papers' note
score 5