Recent work modelling 3D open surfaces train deep neural networks to approximate Unsigned Distance Fields (UDFs) and implicitly represent shapes. To convert this representation to an explicit mesh, they either use computationally expensive methods to mesh a dense point cloud sampling of the surface, or distort the surface by inflating it into a Signed Distance Field (SDF). By contrast, we propose to directly mesh deep UDFs as open surfaces with an extension of marching cubes, by locally detecting surface crossings. Our method is order of magnitude faster than meshing a dense point cloud, and more accurate than inflating open surfaces. Moreover, we make our surface extraction differentiable, and show it can help fit sparse supervision signals.

翻译：最近的工作模拟 3D 开放表面训练深神经网络以近似无符号距离场和隐含代表形状。 要将这个表示法转换为明确的网格, 它们要么使用计算成本昂贵的方法对表面进行稠密点云层取样, 要么通过将表层放大为签名距离场来扭曲表层。 相反, 我们提议直接将深UDF作为开阔表面进行网格, 并在当地探测立方体的延伸。 我们的方法比对稠密点云进行网格的速度要快, 并且比对开放表面进行膨胀更准确。 此外, 我们让地表提取变得不同, 并表明它能帮助适应稀少的监督信号 。