从宽基线立体对学习生成新视角（英文版）

英文标题：Learning to Render Novel Views from Wide-Baseline Stereo Pairs中文摘要：介绍了一种用于单个广角立体图像对的新视角合成的方法，包括 3D 场景重构和外观变化的先验模型，提出了一种多视角变换编码器、图像上的极线采样方案和轻量级交叉注意力渲染器，通过大规模实际数据集的训练，证明了模型学习到了强大的多视角几何先验，大大缩短了渲染时间，并在两个实际数据集上得到了显著的优越性能。英文摘要：We introduce a method for novel view synthesis given only a singlewide-baseline stereo image pair. In this challenging regime, 3D scene pointsare regularly observed only once, requiring prior-based reconstruction of scenegeometry and appearance. We find that existing approa