CLONE: A 3DGS-Based Closed-Loop Differentiable Optimization Framework for Single-Image Normal Estimation
Researchers have developed CLONE, a novel framework for estimating surface normals from single images. This approach utilizes 3D Gaussian Splatting to create a differentiable optimization loop that unifies explicit supervision with generative priors. The system incorporates a differentiable illumination model and a diffusion-inspired refinement network to enhance geometric details and ensure end-to-end differentiability. AI
IMPACT Introduces a new method for geometric modeling in computer vision, potentially improving 3D reconstruction and scene understanding.