Reconstructing 3D Scenes in Native High Dynamic Range

TL;DR

The paper addresses 3D scene reconstruction from native HDR data by showing that traditional LDR-based 3DGS cannot adequately handle HDR radiance. It introduces NH-3DGS, a luminance–chromaticity decomposition that decouples intensity from color, and integrates it with the existing 3D Gaussian Splatting framework to directly optimize from native HDR observations, including RAW Bayer data. Through synthetic and real-world datasets, NH-3DGS achieves state-of-the-art HDR radiance fidelity and efficiency, significantly outperforming multi-exposure HDR methods and previous HDR-NVS approaches while maintaining real-time rendering performance. This work enables professional-grade HDR 3D reconstruction directly from single-exposure HDR captures, reducing capture complexity and preserving radiometric accuracy across a broad dynamic range.

Abstract

High Dynamic Range (HDR) imaging is essential for professional digital media creation, e.g., filmmaking, virtual production, and photorealistic rendering. However, 3D scene reconstruction has primarily focused on Low Dynamic Range (LDR) data, limiting its applicability to professional workflows. Existing approaches that reconstruct HDR scenes from LDR observations rely on multi-exposure fusion or inverse tone-mapping, which increase capture complexity and depend on synthetic supervision. With the recent emergence of cameras that directly capture native HDR data in a single exposure, we present the first method for 3D scene reconstruction that directly models native HDR observations. We propose {\bf Native High dynamic range 3D Gaussian Splatting (NH-3DGS)}, which preserves the full dynamic range throughout the reconstruction pipeline. Our key technical contribution is a novel luminance-chromaticity decomposition of the color representation that enables direct optimization from native HDR camera data. We demonstrate on both synthetic and real multi-view HDR datasets that NH-3DGS significantly outperforms existing methods in reconstruction quality and dynamic range preservation, enabling professional-grade 3D reconstruction directly from native HDR captures. Code and datasets will be made available.

Figures (6)

• Figure 1: Examples of (a) underexposure and (b) overexposure. (c) 3DGS 3dgs exhibits blurring artifacts in dark regions when applying HDR supervision directly. $\Delta t$: Exposure time.
• Figure 2: Elevating the SH order $L$ (3 by default) mitigates blurring artifact while simultaneously inducing additional artifacts.
• Figure 3: The pipeline of NH-3DGS draws inspiration from human visual perception. (a) With a set of HDR training images with corresponding camera poses, NH-3DGS learns a native HDR 3DGS representation. To that end, (b) we reformulate the conventional SH color representation through luminance-chromaticity decomposition. (c) The final HDR color is reconstructed through multiplicative composition of luminance and chromaticity, enabling physically consistent novel view synthesis across the full dynamic range.
• Figure 4: HDR rendering on our collected RAW-4S dataset. 3DGS 3dgs fails to learn a well-calibrated HDR representation, exhibiting spatial blurring in low-illumination regions and chromatic aberrations in neutral highlights (manifesting as greenish/magenta tints).
• Figure 5: Visual comparisons on the Syn-8S dataset.