Enhancing Discontinuous Radiance Fields with Continuous representation
Keywords: Neural Rendering, Neural Geometry Learning, Spiking Neural Network
TL;DR: We propose a dual-branch training strategy that optimizes the density field using both continuous and discontinuous density values.
Abstract: Most NeRF-based methods utilize continuous neural implicit functions (e.g., MLPs) to represent the underlying geometry, which tends to be discontinuous at the interface between the air and the surface. To alleviate this conflict, Spiking NeRF~\cite{liao2024spiking} leverages spiking neurons with a learnable threshold to split the media and object surfaces, achieving promising results but suffering from a lengthy training process. Therefore, in this work, we follow the spirit of Spiking NeRF and extend it to fast grid-based neural fields. Meanwhile, to handle the side effects of spiking neurons, such as holes and floaters, we propose a dual-branch training strategy that optimizes the density field using both continuous and discontinuous density values. The discontinuous branch mirrors Spiking NeRF, with a fixed surrogate gradient, while the continuous branch use full range density values to compute the volume rendering equation, allowing only densities below the threshold to be optimized. Our proposed method can significantly improve both geometry and rendering quality with negligible additional costs. Extensive experimental results show that our method effectively outperforms previous methods on both synthetic and real-world datasets, including solid and transparent objects.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Resubmission: No
Student Author: Yes
Large Language Models: Yes, at the sentence level (e.g., fixing grammar, re-wording sentences)
Submission Number: 11341
Loading