NASA: Neural Articulated Shape Approximation

Efficient representation of articulated objects such as human bodies is an important problem in computer vision and graphics. To efficiently simulate deformation, existing approaches represent 3D objects using polygonal meshes and deform them using skinning techniques. This paper introduces neural articulated shape approximation (NASA), an alternative framework that enables efficient representation of articulated deformable objects using neural indicator functions that are conditioned on pose. Occupancy testing using NASA is straightforward, circumventing the complexity of meshes and the issue of water-tightness. We demonstrate the effectiveness of NASA for 3D tracking applications, and discuss other potential extensions.

💡 Research Summary

The paper introduces Neural Articulated Shape Approximation (NASA), a novel framework for representing articulated deformable objects—such as human bodies—using pose‑conditioned neural occupancy functions instead of traditional polygonal meshes and skinning pipelines. The authors argue that mesh‑based representations suffer from several drawbacks: they require complex spatial acceleration structures (e.g., BVH, octrees) for inside‑outside queries, need to be rebuilt for each deformation, and often lack watertightness, making occupancy testing cumbersome. In contrast, NASA directly models the occupancy function O₍ω₎(x | θ) that maps a 3‑D point x and a pose vector θ to a scalar in