Phase-Aware KANGaussian : Phase-Regularized 3D Gaussian Splatting with Kolmogorov-Arnold Network
Keywords: 3D Gaussian Splatting, Kolmogorov Arnold Network, Phase Regularization, Specular
Abstract: Vanilla 3D Gaussian Splatting struggles with modelling high frequency details, especially in unbounded scenes. Recent works such as Scaffold-GS and Spec-Gaussian have made tremendous improvements to the reconstruction quality of these high frequency details, specifically in synthetic and bounded scenes, but still struggle with unbounded real world scenes. Therefore, we propose Phase-Aware KANGaussian, a model building on these earlier contributions to produce state-of-the-art reconstruction quality for unbounded real world scenes with greatly improved high frequency details. Phase-Aware KANGaussian introduces a novel phase regularization method that optimizes models from low-to-high frequency, dramatically improving the quality of high frequency details. Phase-Aware KANGaussian is also one of the first few papers to integrate a Kolmogorov-Arnold Network (KAN) into the Gaussian Splatting rendering pipeline to verify its performance against the Multilayer Perceptron (MLP). All in all, Phase-Aware KANGaussian has three main contributions: (1) Introduce a Gaussian Splatting model with state-of-the-art performance in modelling real-world unbounded scenes with high frequency details, (2) a novel phase regularization technique to encode spatial representation and lastly, (3) first few to introduce a KAN into the Gaussian Splatting rendering pipeline.
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.
Submission Number: 7201
Loading