Implicit Surface Representations As Layers in Neural Networks

Michalkiewicz, Mateusz; Pontes, Jhony Kaesemodel; Jack, Dominic; Baktashmotlagh, Mahsa; Eriksson, Anders

doi:10.1109/iccv.2019.00484

Cited by 242 publications

(142 citation statements)

References 26 publications

Supporting

Mentioning

142

Contrasting

Order By: Relevance

“…Following the success of CNNs for 2D computer vision problems, many deep learning models have been proposed that can handle 3D data. This includes works on voxel grids [5,9,13,15,27,34,52,53], octrees [46,48], meshes [20,36,49,51], point clouds [12,21,26,33,41,42,50], and implicit functions [35,38]. While they provide effective tools to build predictive models of 3D shapes, e.g.…”

Section: Introductionmentioning

confidence: 99%

Discrete Point Flow Networks for Efficient Point Cloud Generation

Klokov

Boyer

Verbeek³

2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Generative models have proven effective at modeling 3D shapes and their statistical variations. In this paper we investigate their application to point clouds, a 3D shape representation widely used in computer vision for which, however, only few generative models have yet been proposed. We introduce a latent variable model that builds on normalizing flows with affine coupling layers to generate 3D point clouds of an arbitrary size given a latent shape representation. To evaluate its benefits for shape modeling we apply this model for generation, autoencoding, and single-view shape reconstruction tasks. We improve over recent GAN-based models in terms of most metrics that assess generation and autoencoding. Compared to recent work based on continuous flows, our model offers a significant speedup in both training and inference times for similar or better performance. For single-view shape reconstruction we also obtain results on par with state-of-the-art voxel, point cloud, and mesh-based methods.

show abstract

Section: Introductionmentioning

confidence: 99%

Discrete Point Flow Networks for Efficient Point Cloud Generation

Klokov

Boyer

Verbeek³

2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…They are traditionally modeled either as linear combinations of analytic functions or as signed distance grids, which are flexible but memory expensive [55]. Even though the problem of the memory complexity for the grid-based methods is approached by [27,43,57,66,67], they have been outperformed by the recent learning-based continuous representations [2,3,10,12,19,30,39,40,42,45,46,56,62,64]. Furthermore, to improve scalability and representation power, the idea of using local features has been explored in [7,11,41,46,51,52,62].…”

Section: Related Workmentioning

confidence: 99%

LEAP: Learning Articulated Occupancy of People

Mihajlovic

Zhang

Black

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

104

View full text Add to dashboard Cite

“…Implicit Representations: With the introduction of neural implicit functions [12,31,32,36], a number of implicit function-based methods have been proposed to reconstruct human bodies from either images [19,47,48] or point clouds [13]. However, all aforementioned methods treat human body as a rigid object and do not handle registration.…”

Section: Related Workmentioning

confidence: 99%

“…With the advent of neural implicit functions [12,31,32,36], learning-based methods that reconstruct human shapes from point clouds are becoming increasingly accurate [13]. However, most existing neural implicit models treat reconstructed human shapes as static objects and do not provide a way to register such reconstructions to parametric body models.…”

Section: Introductionmentioning

confidence: 99%