Expansive Supervision for Neural Radiance Field

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Weixiang, Xie, Shuzhao, Ge, Shijia, Yao, Wei, Tang, Chen, Wang, Zhi
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912310065364992
author Zhang, Weixiang
Xie, Shuzhao
Ge, Shijia
Yao, Wei
Tang, Chen
Wang, Zhi
author_facet Zhang, Weixiang
Xie, Shuzhao
Ge, Shijia
Yao, Wei
Tang, Chen
Wang, Zhi
contents Neural Radiance Field (NeRF) has achieved remarkable success in creating immersive media representations through its exceptional reconstruction capabilities. However, the computational demands of dense forward passes and volume rendering during training continue to challenge its real-world applications. In this paper, we introduce Expansive Supervision to reduce time and memory costs during NeRF training from the perspective of partial ray selection for supervision. Specifically, we observe that training errors exhibit a long-tail distribution correlated with image content. Based on this observation, our method selectively renders a small but crucial subset of pixels and expands their values to estimate errors across the entire area for each iteration. Compared to conventional supervision, our approach effectively bypasses redundant rendering processes, resulting in substantial reductions in both time and memory consumption. Experimental results demonstrate that integrating Expansive Supervision within existing state-of-the-art acceleration frameworks achieves 52% memory savings and 16% time savings while maintaining comparable visual quality.
format Preprint
id arxiv_https___arxiv_org_abs_2409_08056
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Expansive Supervision for Neural Radiance Field
Zhang, Weixiang
Xie, Shuzhao
Ge, Shijia
Yao, Wei
Tang, Chen
Wang, Zhi
Computer Vision and Pattern Recognition
Neural Radiance Field (NeRF) has achieved remarkable success in creating immersive media representations through its exceptional reconstruction capabilities. However, the computational demands of dense forward passes and volume rendering during training continue to challenge its real-world applications. In this paper, we introduce Expansive Supervision to reduce time and memory costs during NeRF training from the perspective of partial ray selection for supervision. Specifically, we observe that training errors exhibit a long-tail distribution correlated with image content. Based on this observation, our method selectively renders a small but crucial subset of pixels and expands their values to estimate errors across the entire area for each iteration. Compared to conventional supervision, our approach effectively bypasses redundant rendering processes, resulting in substantial reductions in both time and memory consumption. Experimental results demonstrate that integrating Expansive Supervision within existing state-of-the-art acceleration frameworks achieves 52% memory savings and 16% time savings while maintaining comparable visual quality.
title Expansive Supervision for Neural Radiance Field
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2409.08056