High dynamic range (HDR) novel view synthesis (NVS) aims to create photorealistic images from novel viewpoints using HDR imaging techniques. The rendered HDR images capture a wider range of brightness levels containing more details of the scene than normal low dynamic range (LDR) images. Existing HDR NVS methods are mainly based on NeRF. They suffer from long training time and slow inference speed. In this paper, we propose a new framework, High Dynamic Range Gaussian Splatting (HDR-GS), which can efficiently render novel HDR views and reconstruct LDR images with a user input exposure time. Specifically, we design a Dual Dynamic Range (DDR) Gaussian point cloud model that uses spherical harmonics to fit HDR color and employs an MLP-based tone-mapper to render LDR color. The HDR and LDR colors are then fed into two Parallel Differentiable Rasterization (PDR) processes to reconstruct HDR and LDR views. To establish the data foundation for the research of 3D Gaussian splatting-based methods in HDR NVS, we recalibrate the camera parameters and compute the initial positions for Gaussian point clouds. Experiments demonstrate that our HDR-GS surpasses the state-of-the-art NeRF-based method by 3.84 and 1.91 dB on LDR and HDR NVS while enjoying 1000x inference speed and only requiring 6.3% training time. Code and recalibrated data will be publicly available at https://github.com/caiyuanhao1998/HDR-GS . Our project page at https://caiyuanhao1998.github.io/project/HDR-GS/ shows more video and interactive results. An online Gradio demo is also available on HuggingFace at https://huggingface.co/spaces/CaiYuanhao/HDR-GS
Figures & tables
Figure 1: HDR-GS vs. HDR-NeRF. Our HDR-GS achieves better PSNR in dB, SSIM, and LPIPS performance with shorter training time in minutes and faster inference speed in fps.
Figure 2: Comparisons of point clouds (left) and rendered views (right) between the original 3DGS [ 15 ] (top) and our HDR-GS (bottom). (i) 3DGS [ 15 ] renders blurry LDR views when training with images under different exposures. Its point clouds suffer from severe color distortion and can not accurately represent the scene. In addition, 3DGS cannot control the exposure of the rendered images. (ii) Our HDR-GS can not only reconstruct clear HDR images with 3D consistency but also render LDR views with controllable exposure time Δt .
Figure 3: Pipeline of our method. (a) SfM [ 17 ] algorithm is used to recalibrate camera parameters and initialize 3D Gaussians. (b) Dual Dynamic Range Gaussian point clouds use spherical harmonics to model the HDR color. Three MLPs are employed to tone-map the LDR color from the HDR color and user input exposure time. (c) The HDR and LDR colors are fed into two Parallel Differentiable Rasterization to render the HDR and LDR views.
Method
Training
Inference
LDR-OE ( t1,t3,t5 )
LDR-NE ( t2,t4 )
HDR
Time (min)
Speed (fps)
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
NeRF [ 13 ]
405
0.190
13.97
0.555
0.376
14.51
0.522
0.428
—
—
—
3DGS [ 15 ]
38
121
19.46
0.690
0.276
18.97
0.778
0.309
—
—
—
NeRF-W [ 80 ]
437
0.178
29.83
0.936
0.047
29.22
0.927
0.050
—
—
—
HDR-NeRF [ 14 ]
542
0.122
39.07
0.973
0.026
37.53
0.966
0.024
36.40
0.936
0.018
HDR-GS (Ours)
34
126
41.10
0.982
0.011
36.33
0.977
0.016
38.31
0.972
0.013
Table 1: Quantitative results on the synthetic datasets. Metrics are averaged over all scenes. LDR-OE denotes the LDR results with exposure time t1 , t3 , and t5 . LDR-NE denotes the LDR results with exposure time t2 and t4 . HDR denotes the HDR results. HDR-GS yields the best results on all tracks.
Figure 4: LDR visual comparisons on the synthetic scenes. Previous methods introduce unpleasant black spots or render blurry images. Our method controls the exposure better while reconstructing more detailed structures.
Method
LDR-OE ( t1,t3,t5 )
LDR-NE ( t2,t4 )
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
NeRF [ 13 ]
14.95
0.661
0.308
14.44
0.731
0.255
3DGS [ 15 ]
17.19
0.806
0.103
19.50
0.727
0.152
NeRF-W [ 80 ]
28.55
0.927
0.094
28.64
0.923
0.089
HDR-NeRF [ 14 ]
31.63
0.948
0.069
31.43
0.943
0.069
HDR-GS (Ours)
35.47
0.970
0.022
31.66
0.965
0.030
Table 2: Quantitative results on the real datasets. Metrics are averaged across all scenes. LDR-OE represents the LDR results with exposure time t1 , t3 , and t5 . LDR-NE denotes the LDR results with exposure time t2 and t4 . HDR refers to the HDR results. HDR-GS yields the best results on all tracks.
Figure 5: LDR visual comparisons on the real scenes. Previous methods introduce unpleasant black spots or render blurry images. Our method controls the exposure better while reconstructing more detailed structures.
Table 3: Ablations on the synthetic datasets. The PSNR results on HDR, LDR-OE, and LDR-NE are reported.
Figure 6: HDR visual comparisons on the synthetic (upper) and real (lower) scenes. Our method can recover the details in both dark and bright regions while suppressing color distortion. Please zoom in for a better view.
High dynamic range novel view synthesis (HDR-NVS) reconstructs scenes with dynamic details by fusing multi-exposure low dynamic range (LDR) views, yet it struggles to capture ambient illumination-dependent appearance. Implicitly supervising HDR content by constraining tone-mapped results fails in correcting abnormal HDR values, and results in limited gradients for Gaussians in under/over-exposed regions. To this end, we introduce PhysHDR-GS, a physically inspired HDR-NVS framework that models scene appearance via intrinsic reflectance and adjustable ambient illumination. PhysHDR-GS employs a complementary image-exposure (IE) branch and Gaussian-illumination (GI) branch to faithfully reproduce standard camera observations and capture illumination-dependent appearance changes, respectively. During training, the proposed cross-branch HDR consistency loss provides explicit supervision for HDR content, while an illumination-guided gradient scaling strategy mitigates exposure-biased gradient starvation and reduces under-densified representations. Experimental results across realistic and synthetic datasets demonstrate our superiority in reconstructing HDR details (e.g., a PSNR gain of 2.04 dB over HDR-GS), while maintaining real-time rendering speed (up to 76 FPS). Code and models are available at https://huimin-zeng.github.io/PhysHDR-GS/.
Huimin Zeng, Yue Bai, Hailing Wang +1
Department of Electrical and Computer Engineering, Northeastern University · Khoury College of Computer Science, Northeastern University
High dynamic range (HDR) novel view synthesis (NVS) aims to reconstruct HDR scenes from multi-exposure low dynamic range (LDR) images. Existing HDR pipelines heavily rely on known camera poses, well-initialized dense point clouds, and time-consuming per-scene optimization. Current feed-forward alternatives overlook the HDR problem by assuming exposure-invariant appearance. To bridge this gap, we propose InstantHDR, a feed-forward network that initializes 3D HDR scenes from uncalibrated multi-exposure LDR collections in a fast single forward pass. Specifically, we design a geometry-guided appearance modeling for multi-exposure fusion, and a meta-network for generalizable scene-specific tone mapping. Due to the lack of HDR scene data, we build a pre-training dataset, called HDR-Pretrain, for generalizable feed-forward HDR models, featuring 168 Blender-rendered scenes, diverse lighting types, and multiple camera response functions. Comprehensive experiments show that our InstantHDR delivers a single-forward HDR initialization at approximately 700 times the speed of SoTA optimization-based methods, and reaches comparable quality in real settings after lightweight post-optimization while remaining approximately 20 times faster. All code, models, and datasets: https://github.com/Bugjudger/InstantHDR.
Dingqiang Ye, Jiacong Xu, Jianglu Ping +3
Johns Hopkins University, USA · Shenzhen University, China
Single-exposure high dynamic range (HDR) novel view synthesis (NVS) recovers HDR scenes from low dynamic range (LDR) views captured at a single exposure and estimates camera response function (CRF) for LDR re-exposure. Without HDR ground-truth, using single-exposure LDR views as the sole source of direct supervision underconstrains HDR reconstruction and cannot ensure its faithful recovery, motivating the need for learned LDR-to-HDR reconstruction priors. Moreover, single-exposure LDR views supervise CRF only over a limited range, necessitating a learnable prior for reliable re-exposure beyond it. We propose PriorGS, a 3D Gaussian Splatting framework that addresses both challenges by (1) guiding HDR recovery with pseudo-HDR supervision from 2D single-exposure HDR reconstruction methods while anchoring it to input LDR views through a closed reconstruction loop; and (2) regularizing CRF modeling with a learnable prior that guarantees monotonicity and promotes plausible responses. Yet cross-view inconsistencies in independently generated pseudo-HDR targets impair scene reconstruction when HDR is encoded directly as Gaussian's color attribute. We therefore introduce a 3D base-gain HDR representation that factors HDR radiance into an LDR base and an overlaid 3D gain field, promoting robustness to such inconsistencies and enabling 3D HDR appearance editing. Extensive experiments demonstrate PriorGS's substantial superiority over baselines.