iSplat：面向细粒度高斯泼溅的迭代学习
iSplat: Iterative Learning for Fine-Grained Gaussian Splatting
把单次前馈高斯泼溅重建改成 3 次循环的 GRU 迭代优化器，配合基于不确定性的深度搜索空间收缩，可在不增加参数量的前提下显著提升跨数据集新视角合成的泛化；做前馈 3DGS 或深度先验融合的工作可直接借鉴该迭代精化范式。
原文依据 a recurrent GRU-based optimizer that refines both geometry and appearance in a synergistic loop
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期前馈三维高斯泼溅的进展通过单次前向重建场景，展现出显著的效率。然而，这类方法的重建保真度仍落后于传统的基于优化的方法，后者通过漫长的迭代过程逐步修正重建缺陷。本文融合两种范式的优势，提出 iSplat——一个将重建重新表述为包含多次（通常为三次）前向传递的迭代前馈过程的新框架。iSplat 的核心是一个基于 GRU 的循环优化器，它在协同循环中同时精化几何与外观。为应对几何不准确问题，我们提出一种不确定性驱动的深度精化策略，依据每个高斯在上一步估计的不确定性逐步收缩其搜索空间。为进一步改善外观细节，我们设计了一种区域感知增强机制，对重叠区域与非重叠区域分别施加有针对性的多视图与单目特征聚合，以消解歧义。我们在域内（RealEstate10K、ACID）与跨数据集（DTU、ACID）基准上验证了 iSplat 的鲁棒性与泛化能力。iSplat 仅有 42.6M 参数，却在 RealEstate10K 上超越 DepthSplat（354M）（峰值信噪比：27.67 对 27.47 dB）。关键在于，在跨数据集 DTU 基准上，它进一步将峰值信噪比提升 2.88 dB（18.26 对 15.38 dB），展现出卓越的泛化能力。这些结果凸显了迭代精化在克服单样本方法固有局限方面的巨大潜力。我们的代码已发布于 https://github.com/haifengwu205/iSplat。
英文原文摘要 1687 字符
Recent advances in feed-forward 3D Gaussian splatting have demonstrated remarkable efficiency by reconstructing scenes in a single pass. However, the reconstruction fidelity of these methods lags behind that of traditional optimization-based approaches, which gradually correct reconstruction flaws through a lengthy iterative process. In this paper, we leverage the strengths of both paradigms and introduce iSplat, a novel framework that reformulates reconstruction as an iterative feed-forward process involving multiple (typically three) passes. Central to iSplat is a recurrent GRU-based optimizer that refines both geometry and appearance in a synergistic loop. To address geometric inaccuracies, we propose an uncertainty-driven depth refinement strategy that progressively narrows the search space for each Gaussian based on its estimated uncertainty from the previous step. To further improve appearance details, we design a region-aware enhancement mechanism that applies targeted multi-view and monocular feature aggregation to resolve ambiguities in both overlapping and non-overlapping areas. We validate iSplat's robustness and generalization on in-domain (RealEstate10K, ACID) and cross-dataset (DTU, ACID) benchmarks. With only 42.6M parameters, iSplat surpasses DepthSplat (354M) on RealEstate10K (PSNR: 27.67 vs. 27.47 dB). Crucially, on the cross-dataset DTU benchmark, it further boosts the PSNR by 2.88 dB (18.26 vs. 15.38 dB), showcasing exceptional generalization. These results highlight the significant potential of iterative refinement to overcome the inherent limitations of one-shot approaches. Our code is available at https://github.com/haifengwu205/iSplat.