Coarse-to-Fine 3D Face Modeling with Photometric Consistency Optimization

Reconstructing three-dimensional (3D) facial geometry from two-dimensional (2D) images remains a fundamental challenge in computer vision due to its ill-posed nature and vulnerability to occlusions. While classical 3D Morphable Models (3DMMs) and learning-based approaches have shown promise, they often fall short in capturing fine-level facial details. In this work, we propose a coarse-to-fine 3D face reconstruction framework that integrates photometric consistency optimization and a UV position map representation. Our method follows a two-stage training strategy: pre-training on a synthetic dataset to learn general facial structures, followed by fine-tuning on a smaller set of high-resolution 3D scans to improve realism and accuracy. A photometric consistency loss, supervised through left-right paired views, is introduced to further refine texture recovery and geometric fidelity. Experimental results on our custom test set demonstrate that the proposed approach achieves superior detail recovery and accuracy compared to several well-established baseline methods.