GeodesicNVS：面向新视角合成的概率密度测地线流匹配
GeodesicNVS: Probability Density Geodesic Flow Matching for Novel View Synthesis
把扩散的随机噪声到数据路径换成成对视图间的数据到数据确定性流匹配，并用密度测地线约束插值轨迹，可直接用于需要多视角一致性的生成式新视角合成与三维一致生成。
原文依据 We advocate a Data-to-Data Flow Matching framework that learns deterministic transformations between paired views
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论生成建模的最新进展显著提升了新视角合成，但保持跨视角一致性仍然具有挑战性。基于扩散的模型依赖随机的噪声到数据转移，这会模糊确定性结构并导致视角预测不一致。我们主张采用数据到数据流匹配框架，学习成对视图之间的确定性变换，通过显式的数据耦合增强视角一致的合成。在此基础上，我们提出概率密度测地线流匹配（PDG-FM），将插值轨迹与数据流形的基于密度的测地线对齐。为使测地线估计可行，我们采用教师-学生框架，将基于密度的测地线插值蒸馏为高效的环境空间预测器。实验上，我们的方法在 Objaverse 和 GSO30 数据集上超越了基于扩散的基线，展现出更好的结构连贯性和更平滑的跨视角过渡。这些结果凸显了将数据依赖的几何正则化引入确定性流匹配以生成一致新视角的优势。
英文原文摘要 1173 字符
Recent advances in generative modeling have substantially enhanced novel view synthesis, yet maintaining consistency across viewpoints remains challenging. Diffusion-based models rely on stochastic noise-to-data transitions, which obscure deterministic structures and yield inconsistent view predictions. We advocate a Data-to-Data Flow Matching framework that learns deterministic transformations between paired views, enhancing view-consistent synthesis through explicit data coupling. Building on this, we propose Probability Density Geodesic Flow Matching (PDG-FM), which aligns interpolation trajectories with density-based geodesics of a data manifold. To enable tractable geodesic estimation, we employ a teacher-student framework that distills density-based geodesic interpolants into an efficient ambient-space predictor. Empirically, our method surpasses diffusion-based baselines on Objaverse and GSO30 datasets, demonstrating improved structural coherence and smoother transitions across views. These results highlight the advantages of incorporating data-dependent geometric regularization into deterministic flow matching for consistent novel view generation.