Velox：学习 4D 几何与外观的表征
Velox: Learning Representations of 4D Geometry and Appearance
把无序动态点云压缩成一组动态 token，并用 4D 表面解码器与高斯解码器双监督，为视频到 4D 生成、3D 跟踪等任务提供可复用的紧凑 4D 潜变量表示；做动态场景重建或 4D 生成时可直接借用该编码器-双解码器范式。
原文依据 Velox trains an encoder to compress spatiotemporal color point clouds into a set of dynamic tokens.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出了一个用于学习 4D 对象潜变量表征的框架，该表征具有描述性，能够忠实捕捉对象的几何与外观；具有压缩性，有助于提升下游效率；并且易于获取，仅需极少的输入（即一个无序动态点云）即可构建。具体而言，Velox 训练一个编码器，将时空彩色点云压缩为一组动态 token。这些 token 由两个互补的解码器进行监督：4D 表面解码器对随时间变化的表面分布建模，以捕捉几何；高斯解码器将 token 映射为 3D 高斯，以帮助学习外观。为证明该表征的实用性，我们在三个下游任务上对其进行了评估：视频到 4D 生成、3D 跟踪，以及通过图像到 4D 生成实现的布料仿真，并在所有设定下均观察到强劲的性能。
英文原文摘要 887 字符
We introduce a framework for learning latent representations of 4D objects which are descriptive, faithfully capturing object geometry and appearance; compressive, aiding in downstream efficiency; and accessible, requiring minimal input, i.e., an unstructured dynamic point cloud, to construct. Specifically, Velox trains an encoder to compress spatiotemporal color point clouds into a set of dynamic tokens. These tokens are supervised using two complementary decoders: a 4D surface decoder, which models the time-varying surface distribution capturing the geometry; and a Gaussian decoder, which maps the tokens to 3D Gaussians, helping learn appearance. To demonstrate the utility of our representation, we evaluate it across three downstream tasks: video-to-4D generation, 3D tracking, and cloth simulation via image-to-4D generation, and observe strong performances in all settings.