The paper proposes NEMTO, a novel end-to-end neural rendering framework for modeling and synthesizing transparent objects. Key highlights:
NEMTO can handle transparent objects with complex geometry and unknown indices of refraction, which is a challenging problem for traditional physically-based rendering approaches.
The method leverages implicit Signed Distance Functions (SDFs) to represent the object geometry and introduces a refraction-aware Ray Bending Network (RBN) to model the effects of light refraction within the object.
The RBN is more tolerant to geometric inaccuracies compared to traditional physically-based methods, improving the disentanglement of geometry and appearance.
NEMTO can synthesize high-quality novel views and relighting of transparent objects under natural illumination, outperforming existing neural rendering baselines.
The authors provide extensive evaluations on both synthetic and real-world datasets to demonstrate the effectiveness of their approach.
To Another Language
from source content
arxiv.org
Djupare frågor