LTM-NeRF: Embedding 3D Local Tone Mapping in HDR Neural Radiance Field

Area of Science:

• Computer Vision
• Computer Graphics
• Machine Learning

Background:

• Neural Radiance Fields (NeRF) enable novel view synthesis.
• High Dynamic Range NeRF (HDR NeRF) enhances rendering with greater dynamic range.
• Displaying HDR NeRF content on limited dynamic range devices is challenging.

Purpose of the Study:

• To develop a method for recovering HDR NeRF and supporting 3D local tone mapping.
• To enable synthesis of HDR, tone-mapped, and LDR views from multi-view, multi-exposure LDR inputs.
• To ensure compatibility of HDR NeRF content across diverse display devices.

Main Methods:

• Proposed LTM-NeRF method for HDR NeRF reconstruction and 3D local tone mapping.
• Introduced a differentiable Camera Response Function (CRF) module for HDR NeRF reconstruction.
• Developed a Neural Exposure Field (NeEF) for spatially varying exposure time representation and local tone mapping.

Main Results:

• Successfully synthesized HDR views and exposure-varying LDR views accurately.
• Demonstrated natural rendering of locally tone-mapped views.
• Achieved compatibility with various display devices through 3D local tone mapping.

Conclusions:

• LTM-NeRF effectively addresses the challenge of displaying HDR NeRF content on limited dynamic range devices.
• The method allows for versatile synthesis of different view types (HDR, tone-mapped, LDR) with varying exposures.
• LTM-NeRF advances HDR NeRF capabilities for realistic and adaptable visual content generation.