Sparse Statistical Shape Modelling

The provided code is based on the work by A. Gooya et al. [1] which proposes a Gaussian mixture model based approach to training statistical shape models (SSMs). The novel feature of the proposed approach is the application of a symmetric Dirichlet prior on the mixture coefficients to enforce sparsity and search over a continuous space for the optimal number of Gaussian components, to address the common issue of over or under-fitting. Additionally, we provide code to reconstruct surfaces from the unstructured point sets generated, following SSM training.本代码基于A. Gooya等人[1]的研究成果，该工作提出了一种基于高斯混合模型（Gaussian Mixture Model）的统计形状模型（Statistical Shape Models, SSMs）训练方法。该方法的创新之处在于，为混合系数引入对称狄利克雷先验（symmetric Dirichlet prior）以实现稀疏性约束，并在连续空间中搜索最优高斯分量数量，从而解决常见的过拟合与欠拟合问题。此外，本代码还支持在统计形状模型训练完成后，基于生成的非结构化点集重建曲面。