Real-world smoke simultaneously attenuates scene radiance, adds airlight, and destabilizes multi-view appearance consistency, making robust 3D reconstruction particularly difficult. We present SmokeGS-R, a practical pipeline developed for the NTIRE 2026 3D Restoration and Reconstruction Track 2 challenge. The key idea is to decouple geometry recovery from appearance correction: we generate physics-guided pseudo-clean supervision with a refined dark channel prior and guided filtering, train a sharp clean-only 3D Gaussian Splatting source model, and then harmonize its renderings with a donor ensemble using geometric-mean reference aggregation, LAB-space Reinhard transfer, and light Gaussian smoothing. On the official challenge testing leaderboard, the final submission achieved \mbox{PSNR =15.217} and \mbox{SSIM =0.666}. After the public release of RealX3D, we re-evaluated the same frozen result on the seven released challenge scenes without retraining and obtained \mbox{PSNR =15.209}, \mbox{SSIM =0.644}, and \mbox{LPIPS =0.551}, outperforming the strongest official baseline average on the same scenes by +3.68 dB PSNR. These results suggest that a geometry-first reconstruction strategy combined with stable post-render appearance harmonization is an effective recipe for real-world multi-view smoke restoration. The code is available at https://github.com/windrise/3drr_Track2_SmokeGS-R.
Figures & tables
Figure 1: Overview of SmokeGS-R. A refined DCP inversion pipeline produces physics-guided pseudo-clean supervision for a clean-only source 3DGS, while four complementary donor branches provide auxiliary appearance priors. At inference, a five-render geometric-mean reference is constructed and used to harmonize the sharp source rendering in LAB space, followed by light Gaussian smoothing for final smoke-free novel-view restoration.
Method
NVS PSNR ↑
NVS SSIM ↑
NVS LPIPS ↓
3DGS [ 12 ]
9.76
0.499
0.659
SeaThru-NeRF [ 13 ]
7.55
0.464
0.683
Watersplatting [ 14 ]
10.78
0.445
0.723
SeaSplat [ 20 ]
10.42
0.446
0.774
I2-NeRF [ 17 ]
8.40
0.283
0.699
Table 1: Official NVS smoke results reported by RealX3D [ 16 ] . These are the benchmark-level averages reported in the dataset paper.
Method
PSNR ↑
SSIM ↑
LPIPS ↓
3DGS
11.439
0.579
0.631
SeaThru-NeRF
9.013
0.547
0.651
WaterSplatting
9.302
0.426
0.713
SeaSplat
9.277
0.445
0.760
I2-NeRF
7.138
0.255
0.704
Table 2: Averages recomputed from the official public eval_test.json files over the eight released smoke scenes.
Setting
PSNR ↑
SSIM ↑
Development best (#624877)
15.682
0.669
Testing best (#635505)
15.217
0.666
Table 3: Official challenge results for the final SmokeGS-R pipeline. The development score corresponds to our best development-side submission, while the testing score corresponds to the final valid result package used for release.
Method
PSNR ↑
SSIM ↑
LPIPS ↓
3DGS
11.530
0.583
0.627
SeaThru-NeRF
9.133
0.552
0.648
WaterSplatting
9.268
0.424
0.708
SeaSplat
9.001
0.438
0.745
I2-NeRF
7.130
0.256
0.699
SmokeGS-R (ours)
15.209
0.644
0.551
Table 4: Direct comparison on the seven released challenge scenes (28 public test views) covered by our final result package. Official baselines use the organizers’ released eval_test.json values; our row is recomputed against the released reference views. Akikaze is excluded here because it is not part of the frozen final testing package.
Figure 2: Scene-wise PSNR on the seven released challenge scenes. For each scene, we compare SmokeGS-R with the strongest official public baseline on that scene.
Figure 3: Qualitative comparison on the publicly released smoke scenes. Each row corresponds to one scene from the former testing split. We show the released reference view, the official 3DGS baseline, the official SeaThru-NeRF baseline, and our final frozen SmokeGS-R result.
This paper describes our method for Track 2 of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge on smoke-degraded images. In this task, smoke reduces image visibility and weakens the cross-view consistency required by scene optimization and rendering. We address this problem with a multi-stage pipeline consisting of image restoration, dehazing, MLLM-based enhancement, 3DGS-MCMC optimization, and averaging over repeated runs. The main purpose of the pipeline is to improve visibility before rendering while limiting scene-content changes across input views. Experimental results on the challenge benchmark show improved quantitative performance and better visual quality than the provided baselines. The code is available at https://github.com/plbbl/GenSmoke-GS. Our method achieved a ranking of 1 out of 14 participants in Track 2 of the NTIRE 3DRR Challenge, as reported on the official competition website: https://www.codabench.org/competitions/13993/#/results-tab.
Qida Cao, Xinyuan Hu, Changyue Shi +3
School of Computer Science and Technology, Hangzhou Dianzi University · School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen
Reconstructing 3D scenes from smoke-degraded multi-view images is particularly difficult because smoke introduces strong scattering effects, view-dependent appearance changes, and severe degradation of cross-view consistency. To address these issues, we propose a framework that integrates visual priors with efficient 3D scene modeling. We employ Nano-Banana-Pro to enhance smoke-degraded images and provide clearer visual observations for reconstruction and develop Smoke-GS, a medium-aware 3D Gaussian Splatting framework for smoke scene reconstruction and restoration-oriented novel view synthesis. Smoke-GS models the scene using explicit 3D Gaussians and introduces a lightweight view-dependent medium branch to capture direction-dependent appearance variations caused by smoke. Our method preserves the rendering efficiency of 3D Gaussian Splatting while improving robustness to smoke-induced degradation. Results demonstrate the effectiveness of our method for generating consistent and visually clear novel views in challenging smoke environments.
Xinye Zheng, Fei Wang, Yiqi Nie +5
Hefei University of Technology · Institute of Artificial Intelligence, Hefei Comprehensive National Science Center · Anhui University +2
We present Dehaze-then-Splat, a two-stage pipeline for multi-view smoke removal and novel view synthesis developed for Track~2 of the NTIRE 2026 3D Restoration and Reconstruction Challenge. In the first stage, we produce pseudo-clean training images via per-frame generative dehazing using Nano Banana Pro, followed by brightness normalization. In the second stage, we train 3D Gaussian Splatting (3DGS) with physics-informed auxiliary losses -- depth supervision via Pearson correlation with pseudo-depth, dark channel prior regularization, and dual-source gradient matching -- that compensate for cross-view inconsistencies inherent in frame-wise generative processing. We identify a fundamental tension in dehaze-then-reconstruct pipelines: per-image restoration quality does not guarantee multi-view consistency, and such inconsistency manifests as blurred renders and structural instability in downstream 3D reconstruction.Our analysis shows that MCMC-based densification with early stopping, combined with depth and haze-suppression priors, effectively mitigates these artifacts. On the Akikaze validation scene, our pipeline achieves 20.98,dB PSNR and 0.683 SSIM for novel view synthesis, a +1.50,dB improvement over the unregularized baseline.
Boss Chen, Hanqing Wang
The Hong Kong University of Science and Technology (Guangzhou)