NASA: Neural Articulated Shape Approximation

By Timothy Jeruzalski, Boyang Deng, Mohammad Norouzi, John P. Lewis, Geoffrey E. Hinton, Andrea Tagliasacchi. 2019

234 citations recorded by the source corpus.

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.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Geology, 3D Vision & Pose

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Geology

In 3D Vision & Pose

Citations in the corpus

How to Represent Part-Whole Hierarchies in a Neural Network (2021, methodology): “When neural networks are used to represent shape, they generally work much better if they represent the details of a shape relative to its intrinsic coordinate frame rather than relative to a frame based on the camera or the world [Taylor et al., 2007, Deng et al., 2020].”
Decomposing NeRF for Editing via Feature Field Distillation (2022, background): “Conditional [44, 33, 18, 57] and generative models [54, 55, 27] enable a degree of category-speciﬁc decomposition (e.g., human bodyparts) and editing on constrained domains with large datasets.”
A-NeRF: Articulated Neural Radiance Fields for Learning Human Shape, Appearance, and Pose (2021, background): “Closely related to ours is the NASA surface body model [13] that also defines an implicit function as the minimum of individual implicit functions that are rigidly attached to the bones of a skeleton, each conditioned on the entire human pose to model dependencies and learned from 3D scans.”
Animatable Neural Radiance Fields for Modeling Dynamic Human Bodies (2021, background): “[12] combines a set of local implicit functions with human skeletons to represent dynamic humans.”
Convolutional Occupancy Networks (2020, methodology): “dels have been adopted for learning implicit representations from images [23,24,29,41], for encoding texture information [30], for 4D reconstruction [28] as well as for primitive-based reconstruction [10,11,15,32]. Unfortunately, all these methods are limited to 4 S. Peng et al. comparably simple 3D”
D-NeRF: Neural Radiance Fields for Dynamic Scenes (2020, background): “Neural implicit functions have been generalized to articulated objects and non-rigid objects [33, 13] but require full 3D ground-truth supervision.”
Neural Unsigned Distance Fields for Implicit Function Learning (2020, methodology): “IFL methods use either binary occupancies [49, 27, 14, 67, 21, 15] or signed distance functions [53, 13, 50, 36] as shape representation for learning.”
I M Avatar: Implicit Morphable Head Avatars from Videos (2021, background): “a time stamp [57], a learned latent vector [34,56], or a vector from a pre-computed parametric model [14, 18, 29, 42].”