Flow Matching Reinforcement for 3D Mesh Generation via Dynamic Homing Optimization

2026-10-01

AI Summary

This study addresses the limitations of existing reinforcement learning approaches in 3D generation, which typically lack explicit target velocity fields and yield marginal improvements in geometric quality. To overcome these issues, this work proposes the Flow3D-Pro framework, centered on a novel Dynamic Homing Optimization (DHO) method. DHO reformulates negative trajectory optimization as positive sample attraction guidance, introducing for the first time a minimum-cost attraction matching strategy alongside a time-aware dynamic correction mechanism to enable asynchronous online reinforcement learning. Experimental results demonstrate that DHO significantly outperforms mainstream methods such as Direct Preference Optimization (DPO). Furthermore, Flow3D-Pro achieves superior geometric quality compared to state-of-the-art techniques in image-to-3D generation tasks.

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.

Problem

• 3D mesh generation
• flow matching
• reinforcement learning
• geometric quality

Innovation

• Flow Matching Reinforcement Learning
• Dynamic Homing Optimization
• 3D Mesh Generation
• Minimum-Cost Attractive Matching
• Time-Aware Dynamic Correction