Flow3D-OPD: Multi-Teacher On-Policy Distillation for 3D Geometry Generation with Flow-Matching Diffusion Transformer

Researchers posted arXiv:2609.07137v1, a paper introducing Flow3D-OPD, a two-stage post-training framework that applies multi-teacher on-policy distillation to image-to-3D generation built on flow-matching diffusion Transformers. The framework first uses a semi-policy to strengthen the pretrained model and an agentic verifier for 3D geometric quality, then cultivates domain-specialized teacher models via direct preference optimization before consolidating them into a unified student model through on-policy distillation with hard task-routing sampling and gradient accumulation. The authors report consistent improvements across all geometric quality dimensions, with the student surpassing all teacher models on the average metric.

arXiv:2609.07137v1 Announce Type: cross Abstract: Recent image-to-3D generation models built on flow-matching diffusion Transformers DiT can produce high-fidelity meshes, yet their post-training strategy remains largely unexplored. There exist several critical bottlenecks in reinforcement learning: the inherent difficulty of defining comprehensive rewards for 3D geometric quality, and the gradient interference that arises when jointly optimizing heterogeneous objectives. Inspired by the practicability of on-policy distillation OPD in large language models and image generation, we propose \textbf{Flow3D-OPD}, a two-stage post-training framework that introduces multi-teacher distillation into 3D geometry generation. In the first stage, we utilize the semi-policy to enhance the foundational capability of the pretrained model and then design an agentic verifier for 3D geometric quality evaluation. Based on the verifier, we could cultivate domain-specialized teacher models via direct preference optimization DPO . In the second stage, we consolidate heterogeneous expertise into a unified student model through on-policy distillation with hard task-routing sampling and gradient accumulation, which could mitigate the gradient interference in joint optimization. Without relying on elaborate modifications, our straightforward yet effective design achieves consistent improvements across all geometric quality dimensions and surpasses all teacher models in the average metric. Extensive experiments demonstrate that our approach provides an effective paradigm for reinforcement learning in 3D generation.