We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Mip-NeRF RGB-D: Depth Assisted Fast Neural Radiance Fields

Abstract: Neural scene representations, such as Neural Radiance Fields (NeRF), are based on training a multilayer perceptron (MLP) using a set of color images with known poses. An increasing number of devices now produce RGB-D(color + depth) information, which has been shown to be very important for a wide range of tasks. Therefore, the aim of this paper is to investigate what improvements can be made to these promising implicit representations by incorporating depth information with the color images. In particular, the recently proposed Mip-NeRF approach, which uses conical frustums instead of rays for volume rendering, allows one to account for the varying area of a pixel with distance from the camera center. The proposed method additionally models depth uncertainty. This allows to address major limitations of NeRF-based approaches including improving the accuracy of geometry, reduced artifacts, faster training time, and shortened prediction time. Experiments are performed on well-known benchmark scenes, and comparisons show improved accuracy in scene geometry and photometric reconstruction, while reducing the training time by 3 - 5 times.
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Journal reference: Journal of WSCG 2022
DOI: 10.24132/JWSCG.2022.5
Report number: Vol.30., No.1-2, ISSN 1213-6972
Cite as: arXiv:2205.09351 [cs.CV]
  (or arXiv:2205.09351v2 [cs.CV] for this version)

Submission history

From: Arnab Dey [view email]
[v1] Thu, 19 May 2022 07:11:42 GMT (18888kb,D)
[v2] Thu, 9 Jun 2022 11:35:53 GMT (18252kb,D)

Link back to: arXiv, form interface, contact.