Leveraging Intrinsic Components for Few-Shot Neural Radiance Fields in Unconstrained Illumination
Year: 2025
Type: article
Abstract: Neural Radiance Field (NeRF) has demonstrated its efficacy in synthesizing novel view images, and yields promising outcomes under specific conditions: 1) dense, multi-view observations, 2) constrained illumination among input views, and 3) known camera poses. However, these conditions are often unattainable in real-world environments. Especially, when confronted with a limited number of input views and varying illumination conditions, NeRF experiences diminished performance. While some of the previous works have addressed novel view synthesis under unconstrained illumination, their effectiveness still hinges on dense observations. In this paper, we address the challenge of few-shot NeRF under unconstrained illumination. By utilizing intrinsic components that are illumination-invariant across different views, our method enhances NeRF regularization in extreme environments, thereby improving the quality and practicality of synthesis frameworks. We present comprehensive experimental results to prove the efficacy of the proposed method, presenting newly established benchmarks that incorporate multiple viewing directions and unconstrained illumination.
Source: IEEE Access
Institutions: Korea Institute of Science and Technology, Korea University
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
Award: G321574261
OpenAlex ID: W4414270268