GenSmoke-GS: A Multi-Stage Method for Novel View Synthesis from Smoke-Degraded Images Using a Generative Model

Qida Cao; Xinyuan Hu; Changyue Shi; Jiajun Ding; Zhou Yu; Jun Yu

cs.CVApr 3, 2026

GenSmoke-GS: A Multi-Stage Method for Novel View Synthesis from Smoke-Degraded Images Using a Generative Model

Authors: Qida Cao, Xinyuan Hu, Changyue Shi, Jiajun Ding, Zhou Yu, Jun Yu

Organizations: School of Computer Science and Technology, Hangzhou Dianzi University · School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen

Abstract

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.

Figures & tables

Figure 1: Overview of the method. Smoke-degraded images are first restored and dehazed, then enhanced by an MLLM with structural constraints, and finally processed by 3DGS-MCMC. The final output is obtained by averaging the rendered results of repeated runs.

Parameter	Value
MLLM model	GPT-Image-1.5
Optimization iterations	30,000
Number of runs	91

Table 1: Main implementation settings.

Figure 2: Representative qualitative comparison on Futaba, view 0024.

Figure 3: Representative qualitative comparison on Shirohana, view 0027.

Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
3DGS [ 7 ]	11.54	0.597	0.705
I2-NeRF [ 11 ]	7.13	0.257	0.852
SeaSplat [ 13 ]	9.00	0.440	0.827
SeaThru-NeRF [ 9 ]	9.14	0.566	0.767
GenSmoke-GS (Ours)	20.21	0.729	0.446

Table 2: Average quantitative results on the NTIRE 3DRR test set.

Scene	PSNR $\uparrow$	SSIM $\uparrow$
Futaba	20.97	0.807
Hinoki	19.12	0.607
Koharu	20.87	0.783
Midori	20.56	0.791
Natsume	20.76	0.745
Shirohana	17.50	0.569

Table 3: Per-scene results of GenSmoke-GS.

Explore similar work

SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration
Apr 7, 2026Xueming Fu, Lixia Han3D Reconstruction Multi-View 3D Reconstruction
University of Science and Technology of China · Nanjing University of Aeronautics and Astronautics
Dehaze-then-Splat: Generative Dehazing with Physics-Informed 3D Gaussian Splatting for Smoke-Free Novel View Synthesis
Apr 15, 2026Boss Chen, Hanqing WangNovel View Synthesis Multi-View Consistency
The Hong Kong University of Science and Technology (Guangzhou)
3D Smoke Scene Reconstruction Guided by Vision Priors from Multimodal Large Language Models
Apr 7, 2026Xinye Zheng, Fei Wang, Yiqi Nie +53D Reconstruction Multi-View 3D Reconstruction
Hefei University of Technology · Institute of Artificial Intelligence, Hefei Comprehensive National Science Center · Anhui University +2

Apr 7, 2026J/K move · Enter open · S save

SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration

Xueming Fu, Lixia Han

University of Science and Technology of China · Nanjing University of Aeronautics and Astronautics

3D Reconstruction Multi-View 3D Reconstruction 3D Gaussian Splatting Physics-Guided Image Restoration

Open paper AskPDF arXiv

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.

Abstract

Explore similar work

SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration

Dehaze-then-Splat: Generative Dehazing with Physics-Informed 3D Gaussian Splatting for Smoke-Free Novel View Synthesis

3D Smoke Scene Reconstruction Guided by Vision Priors from Multimodal Large Language Models