扩展视角合成 Transformer 的规模
Scaling View Synthesis Transformers
为无几何先验的新视角合成 Transformer 提供了可复用的缩放定律与计算最优设计原则，并指出编码器-解码器结构在同等训练算力下可与 decoder-only 模型同样高效，可直接指导 NVS 模型的算力分配与架构选型。
原文依据 we demonstrate that our encoder-decoder architecture, which we call the Scalable View Synthesis Model (SVSM), scales as effectively as decoder-only models
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论无几何先验的视角合成 Transformer 近期在新视角合成（NVS）中取得了最优性能，超越了依赖显式几何建模的传统方法。然而，决定其随算力扩展的因素仍不明确。本文对视角合成 Transformer 的缩放定律进行系统研究，并推导出训练计算最优 NVS 模型的设计原则。与先前发现相反，我们表明编码器-解码器架构同样可以是计算最优的；我们将此前的负面结果归因于次优的架构选择以及在不等训练算力预算下进行的比较。在多个算力水平上，我们证明所提出的编码器-解码器架构——称为可扩展视角合成模型（SVSM）——的扩展效果与 decoder-only 模型相当，取得了更优的性能-算力帕累托前沿，并在显著降低训练算力的前提下于真实世界 NVS 基准上超越了此前的最优方法。
英文原文摘要 1003 字符
Geometry-free view synthesis transformers have recently achieved state-of-the-art performance in Novel View Synthesis (NVS), outperforming traditional approaches that rely on explicit geometry modeling. Yet the factors governing their scaling with compute remain unclear. We present a systematic study of scaling laws for view synthesis transformers and derive design principles for training compute-optimal NVS models. Contrary to prior findings, we show that encoder-decoder architectures can be compute-optimal; we trace earlier negative results to suboptimal architectural choices and comparisons across unequal training compute budgets. Across several compute levels, we demonstrate that our encoder-decoder architecture, which we call the Scalable View Synthesis Model (SVSM), scales as effectively as decoder-only models, achieves a superior performance-compute Pareto frontier, and surpasses the previous state-of-the-art on real-world NVS benchmarks with substantially reduced training compute.