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

arXiv cs.CV / 4/16/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces “Dehaze-then-Splat,” a two-stage pipeline that first generates pseudo-clean images via per-frame generative dehazing (Nano Banana Pro) and then performs novel view synthesis with 3D Gaussian Splatting (3DGS).
It addresses the key challenge that frame-wise dehazing can harm cross-view consistency, leading to blurred renders and unstable 3D reconstruction, and proposes physics-informed auxiliary losses to mitigate this.
The physics-informed training uses depth supervision aligned with pseudo-depth quality, dark channel prior regularization, and dual-source gradient matching to reduce multi-view inconsistency.
An evaluation on the Akikaze validation scene reports 20.98 dB PSNR and 0.683 SSIM for novel view synthesis, improving by +1.50 dB over an unregularized baseline.
The authors show that MCMC-based densification with early stopping, together with depth and haze-suppression priors, further reduces reconstruction artifacts in smoke removal and 3D rendering.

Abstract

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.