From Prompts to Printable Models: Support-Effective 3D Generation via Offset Direct Preference Optimization

TL;DR

This work tackles the problem of generating 3D models that are inherently printable with minimal support structures. It introduces SEG, a framework that integrates Direct Preference Optimization with an Offset (ODPO) into a diffusion-based 3D generator, guided by a support-structure simulation to penalize high support needs. On Thingi10k-Val and GPT-3DP-Val, SEG consistently achieves substantial reductions in required support volume (NSV) and improved consistency (SEC) over TRELLIS, DPO, and DRO baselines while preserving prompt fidelity. The results highlight SEG's potential to enable more sustainable, efficient digital fabrication by optimizing designs during generation rather than post-hoc adjustments, with planned extensions to other printing modalities and orientations.

Abstract

The transition from digital 3D models to physical objects via 3D printing often requires support structures to prevent overhanging features from collapsing during the fabrication process. While current slicing technologies offer advanced support strategies, they focus on post-processing optimizations rather than addressing the underlying need for support-efficient design during the model generation phase. This paper introduces SEG (\textit{\underline{S}upport-\underline{E}ffective \underline{G}eneration}), a novel framework that integrates Direct Preference Optimization with an Offset (ODPO) into the 3D generation pipeline to directly optimize models for minimal support material usage. By incorporating support structure simulation into the training process, SEG encourages the generation of geometries that inherently require fewer supports, thus reducing material waste and production time. We demonstrate SEG's effectiveness through extensive experiments on two benchmark datasets, Thingi10k-Val and GPT-3DP-Val, showing that SEG significantly outperforms baseline models such as TRELLIS, DPO, and DRO in terms of support volume reduction and printability. Qualitative results further reveal that SEG maintains high fidelity to input prompts while minimizing the need for support structures. Our findings highlight the potential of SEG to transform 3D printing by directly optimizing models during the generative process, paving the way for more sustainable and efficient digital fabrication practices.

Figures (5)

• Figure 1: Overview of the proposed SEG pipeline. The process begins with data sampling from the Thingi10k database, where multi-view images are generated and captioned using Cap3D. The Text-to-Mesh transformation utilizes a sparse flow transformer $G_L$ for model generation. Subsequently, the ODPO training phase incorporates support structure simulation, which includes estimating support volume, identifying risky faces, and assigning rewards based on overlap detection. This integrated approach facilitates the effective generation of 3D models with optimized support requirements.
• Figure 2: An example of the generated mesh featuring risky faces (left) alongside the sliced visualization, highlighting areas that require additional support in blue (right).
• Figure 3: We present the printed results of models generated by the baseline approach (top), which necessitate substantial support structures during the printing process to ensure printability. In contrast, the models produced by our proposed SEG framework (bottom) demonstrate a significantly more efficient requirement for support structures.
• Figure 4: 3D printing process using the Bambu Lab A1 printer for models generated by both the baseline diffusion model and our optimized approach.
• Figure 5: More visualization results generated by TRELLIS xiang2024structured (left) and our model (right). The models are colored in blue, while the risky areas are highlighted in red.