3D Face Reconstruction

3D face reconstruction estimates a person's facial shape, appearance, and often expression or motion from photos, video, or sparse multi-view captures, enabling avatars, performance capture, editing, and analysis. Methods range from 3D morphable models fitted with landmarks and inverse rendering to neural fields and 3D Gaussian splats, with canonical meshes, learned priors, and diffusion used to infer unseen views and fine detail. Current work targets fast few-shot reconstruction, animation and relighting, and consistent meshes, while addressing hair, occlusion, profiles, and changing expressions. A central challenge is balancing identity-faithful geometry and cross-view consistency with photorealism, speed, and generalization from limited input, while ensuring evaluations reflect perceptually important detail and potential bias.