NeRC: Rendering Planar Caustics by Learning Implicit Neural Representations.

Area of Science:

• Computer Graphics
• Computational Imaging
• Photorealistic Rendering

Background:

• Caustics, a complex light transport phenomenon, are crucial for realistic rendering but computationally intensive.
• Existing photon mapping methods struggle with dynamic scenes (moving light/camera) and require substantial resources.
• Rendering dynamic caustics efficiently remains a significant challenge in computer graphics.

Purpose of the Study:

• To develop an efficient method for rendering diverse caustics in scenes with moving cameras and light sources.
• To address the limitations of traditional methods in handling dynamic light transport effects.
• To improve the quality and temporal consistency of rendered caustics.

Main Methods:

• A novel learning-based volume rendering approach utilizing implicit representations.
• Decomposition of appearance into diffuse and specular components using a probabilistic module.
• Integration of implicit signed distance function (SDF) for component weight construction, differing from NeRF.
• Introduction of centering calibration and sine activation for enhanced color prediction.

Main Results:

• The proposed method significantly outperforms baseline approaches in both quantitative and qualitative evaluations.
• Achieves superior performance in rendering caustics for novel views with dynamic light sources.
• Demonstrates improved temporal consistency across frames compared to existing methods.
• Validated on both synthetic and real-world datasets.

Conclusions:

• The learning-based volume rendering approach offers an efficient and effective solution for dynamic caustic rendering.
• The method successfully handles complex light transport effects with moving elements, enhancing realism.
• Presents a significant advancement in rendering temporally consistent and diverse caustics for interactive applications.