SCULPT: Subtractive Composition for 3D Part Generation

2026-08-13

AI Summary

Existing methods for 3D part generation struggle to simultaneously achieve global geometric consistency and high-quality, editable part decomposition: segmentation-based approaches fix the whole shape before partitioning, while additive strategies often produce discontinuous boundaries. This work proposes SCULPT, a novel subtractive part generation framework that operates in a structured 3D latent space, iteratively co-generating parts and the remaining object through joint denoising and sparse overlapping voxel support sets. The approach adaptively determines part counts and effectively eliminates inter-part gaps and material discontinuities. Evaluated on PartObjaverse, SCULPT achieves state-of-the-art geometric quality and excels in assembly-based reconstruction, further enabling fine-grained textured part decomposition from diverse inputs, including real-world images.

Abstract

Part-aware 3D generation aims to create digital assets that are coherent as complete objects while exposing structural parts for editing, material assignment, animation, and reuse. Existing methods impose this structure outside the native generation loop: segmentation-based methods partition an already generated shape, while additive methods synthesize parts from predefined layouts, boxes, or tokens and then reconcile them into a whole. The former preserves the generated geometry but fixes the object before part boundaries are determined; the latter exposes part cardinality but often leaves shared boundaries vulnerable to gaps, interpenetrations, and material discontinuities. In this paper, we propose SCULPT, a framework that addresses these challenges through subtractive composition. Given a complete object represented in a structured 3D latent space, SCULPT iteratively applies a joint split predictor to generate one extracted part together with the remaining object. The predictor performs a coupled denoising process conditioned on both the image and the current 3D state, so the extracted part and updated remainder are generated together rather than reconciled after generation. The joint split predictor processes both outputs on the union of their native sparse 3D supports, allowing neighboring supports to overlap rather than imposing a disjoint voxel partition. The rollout ends when the remainder support becomes empty or reaches a fixed safety cap, allowing the number of generated parts to adapt to each object within that bound. Extensive experiments demonstrate state-of-the-art geometry on PartObjaverse while preserving strong complete-object reconstruction after part assembly. Results on four dataset images, one text-to-image-generated input, and one real-world photograph further show fine-grained textured part decomposition beyond the benchmark.

Problem

• part-aware 3D generation
• subtractive composition
• 3D part decomposition
• structured 3D latent space
• joint split prediction

Innovation

• subtractive composition
• part-aware 3D generation
• joint split predictor
• structured 3D latent space
• adaptive part decomposition