Saved in:
Bibliographic Details
Main Authors: Schnepf, Antoine, Kassab, Karim, Franceschi, Jean-Yves, Caraffa, Laurent, Vasile, Flavian, Mary, Jeremie, Comport, Andrew, Gouet-Brunet, Valérie
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2403.11678
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914799546269696
author Schnepf, Antoine
Kassab, Karim
Franceschi, Jean-Yves
Caraffa, Laurent
Vasile, Flavian
Mary, Jeremie
Comport, Andrew
Gouet-Brunet, Valérie
author_facet Schnepf, Antoine
Kassab, Karim
Franceschi, Jean-Yves
Caraffa, Laurent
Vasile, Flavian
Mary, Jeremie
Comport, Andrew
Gouet-Brunet, Valérie
contents We present a method enabling the scaling of NeRFs to learn a large number of semantically-similar scenes. We combine two techniques to improve the required training time and memory cost per scene. First, we learn a 3D-aware latent space in which we train Tri-Plane scene representations, hence reducing the resolution at which scenes are learned. Moreover, we present a way to share common information across scenes, hence allowing for a reduction of model complexity to learn a particular scene. Our method reduces effective per-scene memory costs by 44% and per-scene time costs by 86% when training 1000 scenes. Our project page can be found at https://3da-ae.github.io .
format Preprint
id arxiv_https___arxiv_org_abs_2403_11678
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Exploring 3D-aware Latent Spaces for Efficiently Learning Numerous Scenes
Schnepf, Antoine
Kassab, Karim
Franceschi, Jean-Yves
Caraffa, Laurent
Vasile, Flavian
Mary, Jeremie
Comport, Andrew
Gouet-Brunet, Valérie
Computer Vision and Pattern Recognition
Machine Learning
We present a method enabling the scaling of NeRFs to learn a large number of semantically-similar scenes. We combine two techniques to improve the required training time and memory cost per scene. First, we learn a 3D-aware latent space in which we train Tri-Plane scene representations, hence reducing the resolution at which scenes are learned. Moreover, we present a way to share common information across scenes, hence allowing for a reduction of model complexity to learn a particular scene. Our method reduces effective per-scene memory costs by 44% and per-scene time costs by 86% when training 1000 scenes. Our project page can be found at https://3da-ae.github.io .
title Exploring 3D-aware Latent Spaces for Efficiently Learning Numerous Scenes
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2403.11678