The field of 3D reconstruction and registration is rapidly advancing, with a focus on developing innovative methods to address the challenges of non-rigid shapes and limited datasets. Recent research has led to the proposal of novel frameworks, such as canonical pose reconstruction models, that enable the transformation of single-view depth images into a canonical form, facilitating shape reconstruction and pose recovery. Additionally, there has been a push towards improving benchmarking tools, with the introduction of modular benchmarking frameworks that allow for the segregation and interchange of fundamental components, enabling a more accurate and efficient evaluation of 3D face reconstruction methods. Furthermore, equivaraint flow matching models have been proposed for point cloud assembly tasks, demonstrating high data efficiency and competitiveness on practical datasets. Noteworthy papers include:

• Canonical Pose Reconstruction from Single Depth Image for 3D Non-rigid Pose Recovery on Limited Datasets, which presents a model that achieves effective results with only a small dataset of approximately 300 samples.
• 3D Face Reconstruction Error Decomposed: A Modular Benchmark for Fair and Fast Method Evaluation, which introduces a toolkit for a Modularized 3D Face reconstruction Benchmark (M3DFB) and presents a computationally efficient approach that penalizes for mesh topology inconsistency.
• Equivariant Flow Matching for Point Cloud Assembly, which presents a novel equivariant solver for assembly tasks based on flow matching models.
• Thickness-aware E(3)-Equivariant 3D Mesh Neural Networks, which proposes a novel framework that effectively integrates the thickness of 3D objects while maintaining computational efficiency.
• NFR: Neural Feature-Guided Non-Rigid Shape Registration, which proposes a novel learning-based framework for 3D shape registration that overcomes the challenges of significant non-rigid deformation and partiality.
• VITON-DRR: Details Retention Virtual Try-on via Non-rigid Registration, which proposes a detail retention virtual try-on method via accurate non-rigid registration for diverse human poses.
• AMBER: Adaptive Mesh Generation by Iterative Mesh Resolution Prediction, which proposes a supervised learning approach to mesh adaptation that generalizes to unseen geometries and consistently outperforms recent baselines.