RewardSDS：通过奖励加权采样对齐分数蒸馏
RewardSDS: Aligning Score Distillation via Reward-Weighted Sampling
AI总结：
本文提出RewardSDS，利用奖励模型的对齐分数加权噪声样本以改进分数蒸馏，并扩展为RewardVSD，在图像、编辑和3D生成任务上显著提升质量与意图对齐。
摘要：
Score Distillation Sampling（SDS，分数蒸馏采样）已成为利用2D扩散先验完成文本到3D生成等任务的有效技术。尽管SDS功能强大，但它难以实现与用户意图的细粒度对齐。为克服这一问题，我们提出RewardSDS，这是一种根据奖励模型给出的对齐分数对噪声样本进行加权的新方法，从而产生加权SDS损失。该损失优先使用来自可产生对齐的高奖励输出的噪声样本的梯度。我们的方法具有广泛适用性，可扩展基于SDS的方法。特别是，我们通过引入RewardVSD展示了其对Variational Score Distillation（VSD，变分分数蒸馏）的适用性。我们在文本到图像、2D编辑和文本到3D生成任务上评估RewardSDS与RewardVSD，结果表明，在衡量生成质量及与期望奖励模型对齐程度的多种指标上，二者相较SDS和VSD均有显著提升，并实现了最先进性能。项目页面位于https://itaychachy.github.io/reward-sds/。
英文摘要：
Score Distillation Sampling (SDS) has emerged as an effective technique for leveraging 2D diffusion priors for tasks such as text-to-3D generation. While powerful, SDS struggles with achieving fine-grained alignment to user intent. To overcome this, we introduce RewardSDS, a novel approach that weights noise samples based on alignment scores from a reward model, producing a weighted SDS loss. This loss prioritizes gradients from noise samples that yield aligned high-reward output. Our approach is broadly applicable and can extend SDS-based methods. In particular, we demonstrate its applicability to Variational Score Distillation (VSD) by introducing RewardVSD. We evaluate RewardSDS and RewardVSD on text-to-image, 2D editing, and text-to-3D generation tasks, showing significant improvements over SDS and VSD on a diverse set of metrics measuring generation quality and alignment to desired reward models, enabling state-of-the-art performance. Project page is available at https://itaychachy.github.io/reward-sds/.
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。
