Restoring 3D scenes with low-light conditions is challenging, and most existing methods depend on precomputed camera poses and scene-specific optimization, which greatly restricts their application to real-world scenarios. To overcome these limitations, we propose Lumos3D, a pose-free single-forward framework for 3D low-light scene restoration. First, we develop a cross-illumination distillation scheme, where a frozen teacher network takes normal-light ground truth images as input to distill accurate geometric information to the student model. Second, we define a Lumos loss to improve the restoration quality of the reconstructed 3D Gaussian space. Trained on a single dataset, Lumos3D performs inference in a purely feed-forward manner, directly restoring illumination and structure from unposed, low-light multi-view images without any per-scene training or optimization. Experiments on real-world datasets demonstrate that Lumos3D achieves competitive restoration results compared to scene-specific methods. Our codes are available at https://github.com/HanzhouLiu/Lumos3D.
Figures & tables
Fig. 1: Architecture overview. Given multi-view low-light images, Lumos3D predicts the restored 3D Gaussian representation in a single-forward way.
Fig. 2: Qualitative comparison of different distillation schemes. Each visualization corresponds to the same scene, with depth on the left and the corresponding RGB image on the right. In the depth maps, blue denotes distant regions and red denotes nearby ones. Distillation on low-light images suffers from illumination ambiguity, whereas distillation on normal-light images yields more accurate and geometrically cleaner depth and relighting results.
Distillation
Lumos losses
Metrics (Average)
Low 1
GT 2
Content
Image
Voxel
PSNR ↑
SSIM ↑
LPIPS ↓
✓
17.93
0.758
0.405
✓
17.47
0.758
0.402
✓
17.47
0.758
0.402
✓
✓
17.59
0.765
0.403
✓
✓
✓
19.41
0.782
0.402
TABLE I: Ablation on distillation targets of the teacher model and Lumos loss variants. Quantitative results are averaged across the LOM dataset using models trained on DL3DV .
Models
Bike
Buu
Chair
Sofa
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
Low-light
Aleth_NeRF [ 4 ]
16.50
0.661
0.481
16.52
0.707
0.418
16.54
0.768
0.536
16.53
0.805
0.408
Luminance-GS [ 5 ]
16.39
0.627
0.520
15.40
0.725
0.436
18.58
0.690
0.634
18.98
0.756
0.472
Lumos3D (Ours)
14.07
0.605
0.432
19.16
0.755
0.420
17.82
0.781
0.565
22.21
0.848
0.346
Over-exposure
TABLE II: Quantitative comparison of different models on the LOM dataset. Best results are highlighted in bold.
Fig. 3: Qualitative comparison of different 3D low-light and over-exposure restoration schemes on the chair and sofa scenes in the LOM dataset.
Novel view synthesis from low-light, noisy, and motion-blurred imagery remains a valuable and challenging task. Current volumetric rendering methods struggle with compound degradation, and sequential 2D preprocessing introduces artifacts due to interdependencies. In this work, we introduce FLED-GS, a fast low-light enhancement and deblurring framework that reformulates 3D scene restoration as an alternating cycle of enhancement and reconstruction. Specifically, FLED-GS inserts several intermediate brightness anchors to enable progressive recovery, preventing noise blow-up from harming deblurring or geometry. Each iteration sharpens inputs with an off-the-shelf 2D deblurrer and then performs noise-aware 3DGS reconstruction that estimates and suppresses noise while producing clean priors for the next level. Experiments show FLED-GS outperforms state-of-the-art LuSh-NeRF, achieving 21× faster training and 11× faster rendering.
Feng Zhang, Jinglong Wang, Ze Li +4
Nanjing University of Posts and Telecommunications, Nanjing. · The Hong Kong University of Science and Technology, Hong Kong. · The Hong Kong Polytechnic University, Hong Kong. +1
This paper presents our approach to the NTIRE 2026 3D Restoration and Reconstruction Challenge (Track 1), which focuses on reconstructing high-quality 3D representations from degraded multi-view inputs. The challenge involves recovering geometrically consistent and photorealistic 3D scenes in extreme low-light environments. To address this task, we propose Extreme Low-light Optimized Gaussian Splatting (ELoG-GS), a robust low-light 3D reconstruction pipeline that integrates learning-based point cloud initialization and luminance-guided color enhancement for stable and photorealistic Gaussian Splatting. Our method incorporates both geometry-aware initialization and photometric adaptation strategies to improve reconstruction fidelity under challenging conditions. Extensive experiments on the NTIRE Track 1 benchmark demonstrate that our approach significantly improves reconstruction quality over the baselines, achieving superior visual fidelity and geometric consistency. The proposed method provides a practical solution for robust 3D reconstruction in real-world degraded scenarios. In the final testing phase, our method achieved a PSNR of 18.6626 and an SSIM of 0.6855 on the official platform leaderboard. Code is available at https://github.com/lyh120/FSGS_EAPGS.
Novel-view synthesis and 3D reconstruction from sparse posed images are central to robotics and AR/VR. Yet, feed-forward 3D Gaussian reconstruction fails under lowlight due to noise, color shifts, and unreliable correspondence. We propose DelowlightSplat, a lowlight-aware feed-forward Gaussian splatting framework for clean novel-view rendering. We build a controllable multi-view lowlight benchmark by degrading only context views while keeping target views clean. We introduce a lightweight Lowlight Adapter for residual enhancement to improve matchability, and couple it with cost-volume-based multi-view inference to directly predict clean 3D Gaussians. Experiments show that DelowlightSplat significantly outperforms previous feed-forward method and two-stage pipeline under lowlight conditions.
Fuzhen Jiang, Zengtian Xie, Zhuoran Li
Hangzhou Dianzi University · Hangzhou, China · Zhuhai College of Science and Technology +1