arXiv 2504.114472025-04-16cs.CV

基于直接偏好优化的扩散蒸馏：高效3D LiDAR场景补全

Diffusion Distillation With Direct Preference Optimization For Efficient 3D LiDAR Scene Completion

Zhejiang University（浙江大学）；Peking University（北京大学）；Zhejiang Green Zhixing Technology co., ltd（浙江绿智行科技有限公司）

An Zhao, Shengyuan Zhang, Ling Yang, Zejian Li, Jiale Wu, Haoran Xu, AnYang Wei, Perry Pengyun GU, Lingyun Sun

备注：: Our code is public available on https://github.com/happyw1nd/DistillationDPO

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对扩散模型在3D LiDAR场景补全中采样慢的问题，提出Distillation-DPO框架，通过生成不同初始噪声的补全场景、以评估指标构建偏好样本对并优化学生模型，实现补全速度超5倍提升且性能更优，为偏好对齐蒸馏提供新方法。

摘要：

扩散模型在3D LiDAR场景补全中的应用受限于其缓慢的采样速度。分数蒸馏可加速扩散采样，但会导致性能下降；而直接策略优化（DPO）后训练则利用偏好数据提升性能。本文提出Distillation-DPO，一种用于LiDAR场景补全的偏好对齐扩散蒸馏新框架。首先，学生模型用不同初始噪声生成配对的补全场景；其次，以LiDAR场景评估指标为偏好，构建胜负样本对，这一构建合理，因多数LiDAR场景指标信息丰富但不可微分，无法直接优化；最后，Distillation-DPO通过利用师生模型在配对补全场景上的分数函数差异优化学生模型，重复该过程直至收敛。大量实验表明，与最先进的LiDAR场景补全扩散模型相比，Distillation-DPO在将补全速度提升超5倍的同时，实现了更高质量的场景补全。据我们所知，该方法首次探索在蒸馏中采用偏好学习，并为偏好对齐蒸馏提供了见解。代码公开于https://github.com/happyw1nd/DistillationDPO。

英文摘要：

The application of diffusion models in 3D LiDAR scene completion is limited due to diffusion's slow sampling speed. Score distillation accelerates diffusion sampling but with performance degradation, while post-training with direct policy optimization (DPO) boosts performance using preference data. This paper proposes Distillation-DPO, a novel diffusion distillation framework for LiDAR scene completion with preference aligment. First, the student model generates paired completion scenes with different initial noises. Second, using LiDAR scene evaluation metrics as preference, we construct winning and losing sample pairs. Such construction is reasonable, since most LiDAR scene metrics are informative but non-differentiable to be optimized directly. Third, Distillation-DPO optimizes the student model by exploiting the difference in score functions between the teacher and student models on the paired completion scenes. Such procedure is repeated until convergence. Extensive experiments demonstrate that, compared to state-of-the-art LiDAR scene completion diffusion models, Distillation-DPO achieves higher-quality scene completion while accelerating the completion speed by more than 5-fold. Our method is the first to explore adopting preference learning in distillation to the best of our knowledge and provide insights into preference-aligned distillation. Our code is public available on https://github.com/happyw1nd/DistillationDPO.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。