Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization

Whereas emerging learning-based scene representations are predominantly evaluated based on image quality metrics such as PSNR, SSIM or LPIPS, only a few investigations focus on the evaluation of geometric accuracy of the underlying model. In contrast to only demonstrating the geometric deviations of...

Full description

Saved in:
Bibliographic Details
Main Authors: D. Haitz, B. Kıvılcım, M. Ulrich, M. Weinmann
Format: Article
Language:English
Published: Copernicus Publications 2024-12-01
Series:The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Online Access:https://isprs-archives.copernicus.org/articles/XLVIII-2-W7-2024/49/2024/isprs-archives-XLVIII-2-W7-2024-49-2024.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850248072853979136
author D. Haitz
B. Kıvılcım
M. Ulrich
M. Weinmann
M. Weinmann
author_facet D. Haitz
B. Kıvılcım
M. Ulrich
M. Weinmann
M. Weinmann
author_sort D. Haitz
collection DOAJ
description Whereas emerging learning-based scene representations are predominantly evaluated based on image quality metrics such as PSNR, SSIM or LPIPS, only a few investigations focus on the evaluation of geometric accuracy of the underlying model. In contrast to only demonstrating the geometric deviations of models for the fully optimized scene model, our work aims at investigating the geometric convergence behavior during the optimization. For this purpose, we analyze the geometric convergence of discretized density fields by leveraging respectively derived point cloud representations for different training steps during the optimization of the scene representation and their comparison based on established point cloud metrics, thereby allowing insights regarding which scene parts are already represented well within the scene representation at a certain time during the optimization. By demonstrating that certain regions reach convergence earlier than other regions in the scene, we provide the motivation regarding future developments on locally-guided optimization approaches to shift the computational burden to the adjustment of regions that still need to converge while leaving converged regions unchanged which might help to further reduce training time and improve the achieved quality.
format Article
id doaj-art-142b756b5ef44ff89e662df60d72c47d
institution OA Journals
issn 1682-1750
2194-9034
language English
publishDate 2024-12-01
publisher Copernicus Publications
record_format Article
series The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
spelling doaj-art-142b756b5ef44ff89e662df60d72c47d2025-08-20T01:58:48ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342024-12-01XLVIII-2-W7-2024495610.5194/isprs-archives-XLVIII-2-W7-2024-49-2024Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal DiscretizationD. Haitz0B. Kıvılcım1M. Ulrich2M. Weinmann3M. Weinmann4Karlsruhe Institute of Technology, Institute of Photogrammetry and Remote Sensing, GermanyKarlsruhe Institute of Technology, Institute of Photogrammetry and Remote Sensing, GermanyKarlsruhe Institute of Technology, Institute of Photogrammetry and Remote Sensing, GermanyKarlsruhe Institute of Technology, Institute of Photogrammetry and Remote Sensing, GermanyDelft University of Technology, Department of Intelligent Systems, The NetherlandsWhereas emerging learning-based scene representations are predominantly evaluated based on image quality metrics such as PSNR, SSIM or LPIPS, only a few investigations focus on the evaluation of geometric accuracy of the underlying model. In contrast to only demonstrating the geometric deviations of models for the fully optimized scene model, our work aims at investigating the geometric convergence behavior during the optimization. For this purpose, we analyze the geometric convergence of discretized density fields by leveraging respectively derived point cloud representations for different training steps during the optimization of the scene representation and their comparison based on established point cloud metrics, thereby allowing insights regarding which scene parts are already represented well within the scene representation at a certain time during the optimization. By demonstrating that certain regions reach convergence earlier than other regions in the scene, we provide the motivation regarding future developments on locally-guided optimization approaches to shift the computational burden to the adjustment of regions that still need to converge while leaving converged regions unchanged which might help to further reduce training time and improve the achieved quality.https://isprs-archives.copernicus.org/articles/XLVIII-2-W7-2024/49/2024/isprs-archives-XLVIII-2-W7-2024-49-2024.pdf
spellingShingle D. Haitz
B. Kıvılcım
M. Ulrich
M. Weinmann
M. Weinmann
Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
title Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
title_full Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
title_fullStr Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
title_full_unstemmed Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
title_short Density-based Geometric Convergence of NeRFs at Training Time: Insights from Spatio-temporal Discretization
title_sort density based geometric convergence of nerfs at training time insights from spatio temporal discretization
url https://isprs-archives.copernicus.org/articles/XLVIII-2-W7-2024/49/2024/isprs-archives-XLVIII-2-W7-2024-49-2024.pdf
work_keys_str_mv AT dhaitz densitybasedgeometricconvergenceofnerfsattrainingtimeinsightsfromspatiotemporaldiscretization
AT bkıvılcım densitybasedgeometricconvergenceofnerfsattrainingtimeinsightsfromspatiotemporaldiscretization
AT mulrich densitybasedgeometricconvergenceofnerfsattrainingtimeinsightsfromspatiotemporaldiscretization
AT mweinmann densitybasedgeometricconvergenceofnerfsattrainingtimeinsightsfromspatiotemporaldiscretization
AT mweinmann densitybasedgeometricconvergenceofnerfsattrainingtimeinsightsfromspatiotemporaldiscretization