Implicit Neural Deformation for Sparse‐View Face Reconstruction

In this work, we present a new method for 3D face reconstruction from sparse-view RGB images. Unlike previous methods which are built upon morphable models (3DMMs) with limited details, leverage an implicit representation to encode rich geometric features. Our overall pipeline consists of two major components, including geometry network, learns deformable neural signed distance function (SDF) as the representation, and rendering render on-surface points SDF match input images via self-supervised optimization. To handle in-the-wild same target different expressions at test time, propose residual latent code effectively expand shape space learned well novel view-switch loss enforce consistency among views. experimental results on several benchmark datasets demonstrate that our approach outperforms alternative baselines achieves superior compared state-of-the-art methods.