GaussHDR: High Dynamic Range Gaussian Splatting via Learning Unified 3D and 2D Local Tone Mapping

CVPR 2025
Jinfeng Liu1, Lingtong Kong2, Bo Li2, Dan Xu1
1The Hong Kong University of Science and Technology, 2vivo Mobile Communication Co., Ltd
Paper Code (Coming soon) Preprocessed Data
SVG mit img laden

3D tone mapping vs. 2D tone mapping. Training with 3D tone mapping often results in inaccurate HDR rendering, while training with 2D tone mapping degrades LDR rendering quality, leading to a higher LDR RMSE metric.

Abstract

High dynamic range (HDR) novel view synthesis (NVS) aims to reconstruct HDR scenes by leveraging multi-view low dynamic range (LDR) images captured at different exposure levels. Current training paradigms with 3D tone mapping often result in unstable HDR reconstruction, while training with 2D tone mapping reduces the model's capacity to fit LDR images. Additionally, the global tone mapper used in existing methods can impede the learning of both HDR and LDR representations. To address these challenges, we present GaussHDR, which unifies 3D and 2D local tone mapping through 3D Gaussian splatting. Specifically, we design a residual local tone mapper for both 3D and 2D tone mapping that accepts an additional context feature as input. We then propose combining the dual LDR rendering results from both 3D and 2D local tone mapping at the loss level. Finally, recognizing that different scenes may exhibit varying balances between the dual results, we introduce uncertainty learning and use the uncertainties for adaptive modulation. Extensive experiments demonstrate that GaussHDR significantly outperforms state-of-the-art methods in both synthetic and real-world scenarios.

Method Overview

SVG mit img laden

The overview of GaussHDR:

  1. (a) We assign each 3D Gaussian with a context feature for 3D local tone mapping and uncertainty prediction. The HDR irradiance, context feature, LDR color and uncertainty are simultaneously rendered onto the image plane.
  2. (b) We perform 2D local tone mapping on the rendered HDR image and feature map and predict the uncertainty.
  3. (c) We combine the dual LDR rendering results under 3D and 2D local tone mapping at the loss level and utilize their uncertainties for adaptive modulation.

Demo Videos

We provide demo videos of novel HDR and LDR renderings for four scenes.

HDR Comparisons

HDR-GS vs. Ours-GS

We reimplement HDR-GS for fair comparison. HDR GTs of synthetic data are not used for supervision.

HDR-GS Ours-GS HDR-GS Ours-GS HDR-GS Ours-GS HDR-GS Ours-GS HDR-GS Ours-GS

HDR-Scaffold-GS vs. Ours-Scaffold-GS

We replace the scene representation in HDR-GS from 3DGS to Scaffold-GS to establish a new baseline for comparison.

HDR-Scaffold-GS Ours-Scaffold-GS HDR-Scaffold-GS Ours-Scaffold-GS HDR-Scaffold-GS Ours-Scaffold-GS HDR-Scaffold-GS Ours-Scaffold-GS HDR-Scaffold-GS Ours-Scaffold-GS

LDR Comparisons

We present the LDR qualitative comparisons through error maps, since it is difficult to distinguish the differences between LDR images with the naked eye. The following results can be also found in our supplementary material.

SVG mit img laden

BibTeX

@inproceedings{gausshdr,
  author    = {Jinfeng Liu and Lingtong Kong and Bo Li and Dan Xu},
  title     = {GaussHDR: High Dynamic Range Gaussian Splatting via Learning Unified 3D and 2D Local Tone Mapping},
  booktitle = {CVPR}, 
  year      = {2025},
}