Search papers, labs, and topics across Lattice.
The paper introduces Metric-Phase Fields (MPFs), a novel implicit representation that decouples metric proximity from topological phase to improve thin-structure reconstruction from unoriented point clouds. MPFs learn an unsigned metric field and a smooth phase field, coupled via a gated-metric formulation with residual phase injection, resulting in a signed implicit function with stable gradients. Experiments on synthetic and scanned data demonstrate that MPFs outperform SDF-based methods in preserving thin structures and UDF-based methods in training robustness and surface extraction reliability.
Achieve high-fidelity 3D reconstruction of thin structures from point clouds by decoupling metric distance and topological phase, sidestepping the limitations of both SDFs and UDFs.
Neural Signed Distance Functions (SDFs) excel at reconstructing watertight manifolds but fail on thin structures and open boundaries due to strict inside--outside constraints. Conversely, Unsigned Distance Fields (UDFs) accommodate general geometries but suffer from gradient singularities at the zero-level set, hindering optimization and extraction. We introduce Metric--Phase Fields (MPFs), a decoupled implicit representation that separates metric proximity from topological phase. Given an unoriented point cloud, MPFs learn (i) an unsigned metric field $r$ and (ii) a smooth phase field $胃$, for which we derive a bounded phase indicator $P=\tanh(尾胃)$ that provides soft inside--outside cues where they are meaningful. We couple the two fields via a gated-metric formulation with a residual phase injection to obtain a signed implicit function with stable near-surface gradients. The phase coefficient $尾$ is learnable, allowing MPFs to adaptively control the sharpness of the phase transition and the degree of saturation of the soft sign indicator. Experiments on both synthetic and scanned thin-shell and thin-plate shapes demonstrate that MPFs preserve thin and layered structures more faithfully than recent SDF-based methods, while also enabling more robust training and more reliable surface extraction than UDF-based approaches. Check out \href{https://github.com/JIAYI-Scarlett/ICML2026-MPF}{MPFs-GitHub} for source code and test models.
