Position forcing improves 3d shape generation quality in single-stage models
Position Forcing: Self-Conditioning 3D Generation
Computer Vision and Pattern Recognition
Summary
Generating 3D shapes can be tricky when a model doesn't clearly know where each part belongs. The authors noticed that even without explicit position information, these models keep some hints about where parts are located. They created Position Forcing, a technique that figures out and feeds back position hints during the shape-building process, helping the model place parts more accurately over time. This leads to better 3D shape creation without needing extra steps to find positions first.
What this means in practice
- •For 3d graphics developers: Generate higher-quality 3D shapes in a single step for use in games, simulations, or visual effects.
- •For augmented reality app teams: Create more precise 3D object models on the fly to enhance AR experiences without complex preprocessing pipelines.
Authors
Ziheng Ouyang, Zeqiang Lai, Jiarui Chen, Jiangshan Wang, Yuhao Wan, Jingbo Gong, Xiangyu Yue, Hengshuang Zhao, Qibin Hou, Chunchao Guo
Abstract
Recent single-stage 3D generative models commonly adopt VecSet representations, encoding 3D shapes as unordered sets of latent tokens. However, compared with two-stage methods that provide explicit positional guidance, these models must implicitly infer token positions throughout denoising, limiting their generation quality. We observe that, despite the absence of explicit positional conditioning, VecSet tokens retain recoverable spatial correspondences. Building on this observation, we propose Position Forcing, a position-based self-conditioning framework. During denoising, Position Forcing recovers token positions from the current clean latent estimate, quantizes them at progressively finer resolutions according to the denoising stage, and feeds the resulting positional encodings back into the diffusion Transformer. This progressively refined positional feedback provides spatial guidance at a granularity appropriate to each denoising stage, guiding shape generation along a coarse-to-fine trajectory and substantially improving generation quality without a separate position generation stage. Experiments demonstrate that Position Forcing achieves strong performance among single-stage 3D generative methods and outperforms several competitive multi-stage approaches.