PartUV: Part-Based UV Unwrapping of 3D Meshes

TL;DR

PartUV presents a part-based UV unwrapping framework that leverages semantic part priors from PartField in a top-down recursive tree search to produce compact, part-aligned charts with low distortion. By combining two geometry-based heuristics (Normal and Merge) with ABF++ flattening and robust runtime optimizations, PartUV achieves significantly fewer charts and shorter seams than traditional tools while maintaining competitive distortion across diverse datasets, including AI-generated and CAD meshes. The approach enables practical applications such as part-aware multi-atlas packing and texture editing, and demonstrates strong robustness via 95th percentile distortion analyses. Overall, PartUV advances UV unwrapping by integrating semantic priors with geometry-driven cuts to deliver fast, scalable, and semantically meaningful UV maps for complex and noisy meshes.

Abstract

UV unwrapping flattens 3D surfaces to 2D with minimal distortion, often requiring the complex surface to be decomposed into multiple charts. Although extensively studied, existing UV unwrapping methods frequently struggle with AI-generated meshes, which are typically noisy, bumpy, and poorly conditioned. These methods often produce highly fragmented charts and suboptimal boundaries, introducing artifacts and hindering downstream tasks. We introduce PartUV, a part-based UV unwrapping pipeline that generates significantly fewer, part-aligned charts while maintaining low distortion. Built on top of a recent learning-based part decomposition method PartField, PartUV combines high-level semantic part decomposition with novel geometric heuristics in a top-down recursive framework. It ensures each chart's distortion remains below a user-specified threshold while minimizing the total number of charts. The pipeline integrates and extends parameterization and packing algorithms, incorporates dedicated handling of non-manifold and degenerate meshes, and is extensively parallelized for efficiency. Evaluated across four diverse datasets, including man-made, CAD, AI-generated, and Common Shapes, PartUV outperforms existing tools and recent neural methods in chart count and seam length, achieves comparable distortion, exhibits high success rates on challenging meshes, and enables new applications like part-specific multi-tiles packing. Our project page is at https://www.zhaoningwang.com/PartUV.

Figures (14)

• Figure 1: Pipeline of PartUV. Given a mesh $\mathcal{M}$, we first leverage the learning-based method PartField to predict a part-aware feature field. By applying a clustering algorithm to this field, we obtain a hierarchical part tree $\mathcal{T}$ for the input mesh $\mathcal{M}$. We then recursively traverse the part tree $\mathcal{T}$ starting from the root node. For each visited node $\mathcal{P}$, we apply two novel geometry-based strategies to segment the corresponding part mesh into multiple sets of charts $\mathscr{C}$. Each chart in the set is then flattened using the ABF algorithm sheffer2005abf++, and its distortion is evaluated. If the distortion exceeds a user-specified threshold $\tau$, we recursively traverse the left and right children of the part tree $\mathcal{T}$. Otherwise, we adopt the segmented charts and their corresponding UV mappings for the part mesh.
• Figure 2: Quantitative comparison between Blender blender, xatlas xatlas, and PartUV. Unlike Blender and xatlas, which rely solely on local geometric properties for UV unwrapping, PartUV integrates high-level semantic priors with low-level geometric heuristics, enabling part-based chart decomposition. As a result, it produces significantly fewer charts, with boundaries that better align with semantic parts.
• Figure 3: Open3D not only suffers from a limited success rate but also produces results with large distortion, which may reduce their practical utility.
• Figure 4: Nuvo and OptCuts incur significantly longer optimisation times for each shape. Nuvo’s results also exhibit large distortion.
• Figure 5: Texture-map comparison among Blender, xatlas, and our PartUV.