Ref-GeNVS：面向镜面场景的免训练生成式新视角合成

Translation Summary

Okay, so the task is to translate this English text into Simplified Chinese, keeping paragraph breaks, URLs, and eliminating all commentary. The subject matter is right in my wheelhouse: Computer Vision, specifically novel view synthesis, and the application of diffusion models to 3D reconstruction problems. This sounds interesting.

First, I'll go through it sentence by sentence. I need to get the gist of each sentence, noting key technical terms. For the first sentence: "We propose Ref-GeNVS, a training-free, reflection-aware method for generative novel view synthesis (NVS) in mirror scenes." I need to preserve "Ref-GeNVS," and translate "training-free," "reflection-aware," "generative novel view synthesis (NVS)," and "mirror scenes" accurately. For the second sentence: "Existing multi-view diffusion models often fail to recognize the mirror in the scene and cannot exploit reflected content for scene generation," the challenge is to correctly translate “multi-view diffusion models," "fail to recognize," and "exploit reflected content."

The third sentence, "To fix this issue without additional training, our key idea is to treat a mirror image as two complementary views," is important because it describes the core concept. I need to accurately translate the phrase "treat a mirror image as two complementary views." I'm thinking about "mirror image" as the image containing the reflection, which means it really is two viewpoints: the real scene as captured and the "virtual" view behind the mirror. This "virtual" view is like a camera placed behind the mirror.

For the fourth sentence: "From input images, we estimate the mirror plane and reflect camera poses to form virtual views." - I'll want to choose the right verb for reflecting camera poses – like "mirror" or "transform." And for the fifth: "Based on this virtual view setup, we propose a two-stage generation method consisting of Mirror-gated attention and Reflection injection, which enables reflection-consistent NVS by explicitly leveraging reflection relationships in a multi-view diffusion model." It's important to keep the technical terms like "Mirror-gated attention" and "Reflection injection" consistent. And translate that "reflection-consistent NVS" accurately. The sixth sentence: "Ref-GeNVS inherits the strong generalizability of the multi-view diffusion backbone, while it does not require finetuning," the key is "multi-view diffusion backbone," and the last sentence "On synthetic and real scenes including mirrors, Ref-GeNVS outperforms recent generative NVS methods by generating reflection-consistent and contextually coherent novel views, revealing scene structure visible only through mirrors. Project page: https://kim-geonu.github.io/Ref-GeNVS/” - I need a solid translation for "outperforms," and "reflection-consistent and contextually coherent" and of course, I must maintain the URL.

After careful consideration of the technical aspects, I draft the translated text. I review the translation to ensure all key concepts are accurate and the meaning clear. I replace more precise technical expressions. Finally, I will check the translation against the constraints: simplified Chinese, paragraph preservation, URL preservation, and no added commentary.

我们提出了 Ref-GeNVS，一种用于镜面场景中生成式新视角合成（NVS）的免训练、反射感知方法。现有的多视角扩散模型通常难以识别场景中的镜子，且无法利用反射内容进行场景生成。为了在无需额外训练的情况下解决这一问题，我们的核心思想是将镜面图像视为两个互补的视角。我们从输入图像中估计镜面平面，并通过对相机位姿进行镜像变换来构建虚拟视角。基于这种虚拟视角设置，我们提出了一种由镜面门控注意力（Mirror-gated attention）和反射注入（Reflection injection）组成的两阶段生成方法，通过在多视角扩散模型中显式利用反射关系，实现了具有反射一致性的 NVS。Ref-GeNVS 继承了多视角扩散骨干网络的强大泛化能力，同时无需微调。在包含镜子的合成与真实场景中，Ref-GeNVS 通过生成反射一致且上下文连贯的新视角，揭示了仅在镜中可见的场景结构，性能优于近期的生成式 NVS 方法。项目主页：https://kim-geonu.github.io/Ref-GeNVS/