Flow matching method improves 3d mesh generation quality
Flow Matching Reinforcement for 3D Mesh Generation via Dynamic Homing Optimization
Computer Vision and Pattern Recognition
Summary
Generating 3D shapes like meshes often relies on techniques developed for 2D images, which may not work well for complex 3D data. The authors propose a new reinforcement learning approach called Dynamic Homing Optimization that directs the process toward better 3D shapes by pairing each less-desired sample with a specific better target and guiding it dynamically. This method helps create higher-quality 3D meshes compared to previous strategies. They demonstrate their approach within a system called Flow3D-Pro that produces better 3D geometry than existing mesh generation tools.
What this means in practice
- •For 3d model developers: Generate higher-quality 3D meshes from images using improved flow matching and reinforcement learning.
- •For computer graphics engineers: Enhance mesh generation pipelines by integrating Dynamic Homing Optimization for better geometric accuracy in 3D reconstruction tasks.
Authors
Zhen Zhou, Zhiwei Ning, Puhua Jiang, Sheng Zhang, Yifei Tang, Jie Yang, Xintong Han, Wei Liu, Chunchao Guo
Abstract
Flow matching is central to 3D generation, yet in practice its reinforcement learning (RL) methods are largely adapted from 2D visual generation. Representative DPO-, GRPO-, and NFT-style objectives, when applied to negative trajectories, mainly steer predicted velocities away from the corresponding directions without explicitly specifying a target velocity field toward preferred samples. In 3D generation, constrained by pretrained model capabilities, rollout diversity, and reward-distribution complexity, directly applying these RL methods yields limited gains in geometric quality. We introduce a forward-process RL method \textbf{Dynamic Homing Optimization (DHO)}, which reformulates negative-trajectory optimization as positive-sample attraction-guided dynamic homing. Specifically, Minimum-Cost Attractive Matching (MAM) assigns each negative sample a distinct positive target, and Time-Aware Dynamic Correction (TDC) then redirects its trajectory toward the target using a remaining-time-aware corrective velocity. Building on asynchronous online DHO, we develop \textbf{Flow3D-Pro}, an image-to-3D geometry generation framework. Experiments show that DHO outperforms representative DPO-, GRPO-, and NFT-style objectives in 3D generation, while Flow3D-Pro produces higher-quality 3D geometry than existing mesh generation methods.