FullPart: Generating each 3D Part at Full Resolution

TL;DR

This work addresses the challenge of detailed, coherent part-based 3D generation by marrying implicit layout modeling with explicit per-part full-resolution grids. The FullPart framework first derives bounding-box layouts via an implicit diffusion process and then generates each part inside its own high-resolution voxel grid, solving the resolution bottleneck of global grids. A center-corner encoding strategy ensures cross-part alignment across varying part sizes, enabling smooth stitching and global coherence. To support learning, PartVerse-XL provides a large-scale, high-quality dataset with 320K parts and part-aware captions. Experiments show state-of-the-art results in part fidelity and occlusion handling, plus practical interactive editing capabilities.

Abstract

Part-based 3D generation holds great potential for various applications. Previous part generators that represent parts using implicit vector-set tokens often suffer from insufficient geometric details. Another line of work adopts an explicit voxel representation but shares a global voxel grid among all parts; this often causes small parts to occupy too few voxels, leading to degraded quality. In this paper, we propose FullPart, a novel framework that combines both implicit and explicit paradigms. It first derives the bounding box layout through an implicit box vector-set diffusion process, a task that implicit diffusion handles effectively since box tokens contain little geometric detail. Then, it generates detailed parts, each within its own fixed full-resolution voxel grid. Instead of sharing a global low-resolution space, each part in our method - even small ones - is generated at full resolution, enabling the synthesis of intricate details. We further introduce a center-point encoding strategy to address the misalignment issue when exchanging information between parts of different actual sizes, thereby maintaining global coherence. Moreover, to tackle the scarcity of reliable part data, we present PartVerse-XL, the largest human-annotated 3D part dataset to date with 40K objects and 320K parts. Extensive experiments demonstrate that FullPart achieves state-of-the-art results in 3D part generation. We will release all code, data, and model to benefit future research in 3D part generation.

Figures (8)

• Figure 1: FullPart achieves high-quality part-based 3D generation.
• Figure 2: FullPart framework. FullPart comprises three sequential stages: (a) layout generation using implicit vecset diffusion, (b) generating each part at a full-resolution grid with explicit voxel representation, and (c) refining coarse part structures to texture meshes.
• Figure 3: Illustration of our 3D part representation. Our model generates each part at isolated full resolution (c), which contains more fine details than the previous sharing global voxel grid strategy (b). Also, tokens from different parts represent varying spatial extents, e.g., head and body in (d).
• Figure 4: Comparison with state-of-the-art 3D Part generators. Our method can generate more detailed and reasonably divided parts.
• Figure 5: Comparison with the state-of-the-art 3D generators.