X-Part: High Fidelity And Structure Coherent Shape Decomposition And Completion

Generating 3D shapes at part level is pivotal for downstream applications such as mesh retopology, UV mapping, and 3D printing. However, existing part-based generation methods often lack sufficient controllability and produce semantically inconsistent decompositions. To this end, we introduce X -Part, a diffusion-based method designed to decompose a holistic 3D object into semantically meaningful and structurally coherent parts with high geometric fidelity.

X -Part exploits bounding boxes as prompts for part generation and injects point-wise semantic features for meaningful decomposition. Furthermore, we design a pipeline for interactive part editing. Extensive experimental results show that X -Part significantly advances the state-of-the-art in both part shape quality and semantic correctness. This work establishes a new paradigm for creating productionready, editable, and structurally sound 3D assets. Codes will be released for public research.