SciTePress - Affiliation Profile

Research.Publish.Connect.

Title:
ISBN:
Year:
Acronym:
Subject:

Advanced Search Proceedings Search

If you're looking for an exact phrase use quotation marks on text fields.

Title:
Author:
Affiliation:
Subject:

Advanced Search Papers Search

If you're looking for an exact phrase use quotation marks on text fields.

Name:
Affiliation:
Country:
Conference:
Subject:

Advanced Search Authors Search

If you're looking for an exact phrase use quotation marks on text fields.

Name:
Country:
Subject:

Advanced Search Affiliations Search

If you're looking for an exact phrase use quotation marks on text fields.

Proceedings

Proceedings Search

Title:
ISBN:
Year:
Acronym:
Subject:

Advanced Search Proceedings Search

If you're looking for an exact phrase use quotation marks on text fields.

Papers

Papers Search

Title:
Author:
Affiliation:
Subject:

Advanced Search Papers Search

If you're looking for an exact phrase use quotation marks on text fields.

Authors

Authors Search

Name:
Affiliation:
Country:
Conference:
Subject:

Advanced Search Authors Search

If you're looking for an exact phrase use quotation marks on text fields.

Advanced Search

Affiliation

German Research Center for Artificial Intelligence, Stuhlsatzenhausweg 3 (Campus D3 2), Saarbr ücken, Germany

Context (please use complete words only)

Field

Sort by

1 Papers

HD-VoxelFlex: Flexible High-Definition Voxel Grid Representation

Topics: Image-Based Modeling and 3D Reconstruction; Features Extraction; 3D Deep Learning

Authors: Igor Vozniak , Pavel Astreika , Philipp Müller , Nils Lipp , Christian Müller and Philipp Slusallek

Affiliation: German Research Center for Artificial Intelligence, Stuhlsatzenhausweg 3 (Campus D3 2), Saarbr ücken, Germany

Digital Object Identifier: 10.5220/0012374800003660

Proceedings of VISAPP 2024, ISBN: 978-989-758-679-8

Quick Abstract

Voxel grids are an effective means to represent 3D data, as they accurately preserve spatial relations. However, the inherent sparseness of voxel grid representations leads to significant memory consumption in deep learning architectures, in particular for high-resolution (HD) inputs. As a result, current state-of-the-art approaches to the reconstruction of 3D data tend to avoid voxel grid inputs. In this work, we propose HD-VoxelFlex, a novel 3D CNN architecture that can be flexibly applied to HD voxel grids with only moderate increase in training parameters and memory consumption. HD-VoxelFlex introduces three architectural novelties. First, to improve the models’ generalizability, we introduce a random shuffling layer. Second, to reduce information loss, we introduce a novel reducing skip connection layer. Third, to improve modelling of local structure that is crucial for HD inputs, we incorporate a kNN distance mask as input. We combine these novelties with a “bag of tricks” identified in a comprehensive literature review. Based on these novelties we propose six novel building blocks for our encoder-decoder HD-VoxelFlex architecture. In evaluations on the ModelNet10/40 and PCN datasets, HD-VoxelFlex outperforms the state-of-the-art in all point cloud reconstruction metrics. We show that HD-VoxelFlex is able to process high-definition (128 3 , 192 3 ) voxel grid inputs at much lower memory consumption than previous approaches. Furthermore, we show that HD-VoxelFlex, without additional fine-tuning, demonstrates competitive performance in the classification task, proving its generalization ability. As such, our results underline the neglected potential of voxel grid input for deep learning architectures.