Guiding Image-to-3D Generation with Test-Time Partial Observations

While single-image 3D generative models produce visually appealing assets, their underlying geometry is often under-constrained. This paper presents a training-free test-time guidance framework that injects partial geometric observations directly into pretrained generators. By formulating a ray-consistent observation likelihood over the model's occupancy field—accounting for both surface presence and free-space evidence—the method substantially improves geometric fidelity on models like SAM 3D without parameter updates.