Jin, H., Jian, M., Ding, D. and Yu, H. (2025) Self-Supervised Face Deocclusion via 3-D Face Reconstruction with Outlier Segmentation. IEEE Transactions on Human-Machine Systems, 55, 746-755.
https://doi.org/10.1109/thms.2025.3585780

has been cited by the following article:

• TITLE: 3D Face Reconstruction with Implicit Neural Representation and Multi-Scale Feature Fusion

  AUTHORS: Danni Peng, Guoliang Wei, Yuhua Ai

  KEYWORDS: 3D Morphable Models, Implicit Neural Representations, Dense Atrous Convolution, Spatial Regularization

  JOURNAL NAME: Open Journal of Applied Sciences, Vol.15 No.12, December 18, 2025

  ABSTRACT: In computer vision and graphics applications, the precise representation of 3D faces is of critical importance. This paper proposes a novel 3D deformable face model that learns complex continuous spaces through implicit representation. Firstly, multi-scale context features are extracted from the input image by using a dense dilated convolution branch, capturing both global semantics and local geometric details. Then, through position encoding and gated fusion, an adaptive mapping between image features and 3D spatial coordinates is achieved. To enhance the implicit decoding capability, local expert decoders are constructed, and spatial regularization constraints are introduced to ensure the local continuity and geometric smoothness of the implicit field. Experiments show that this method performs well on the FaceScape dataset, with a chamfer distance of 0.553 and an F-score of 93.74. It also demonstrates high-fidelity details in 3D face reconstruction when compared with multiple classic algorithms.