CLIP-Guided Data Augmentation for Night-Time Image Dehazing

Xining Ge; Weijun Yuan; Gengjia Chang; Xuyang Li; Shuhong Liu

cs.CVApr 7, 2026

CLIP-Guided Data Augmentation for Night-Time Image Dehazing

Authors: Xining Ge, Weijun Yuan, Gengjia Chang, Xuyang Li, Shuhong Liu

Organizations: Hangzhou Dianzi University · Jinan University · Hefei University of Technology · Wuhan University ·

Abstract

Nighttime image dehazing faces a more complex degradation pattern than its daytime counterpart, as haze scattering couples with low illumination, non-uniform lighting, and strong light interference. Under limited supervision, this complexity aggravates domain drift and training instability, since target-domain samples are scarce while naively introducing external data may weaken adaptation due to distribution mismatch. This paper presents our solution to the NTIRE 2026 Night Time Image Dehazing Challenge, built as a unified framework that integrates domain-aligned data construction, stage-wise training, and inference-time enhancement. Specifically, a pre-trained CLIP visual encoder screens candidate external samples by similarity to construct training data closer to the target domain. NAFNet is then trained in two stages, first adapting to the target domain and then expanding to broader degradation patterns. At inference time, TLC, x8 self-ensemble, and weighted snapshot fusion are combined to improve output stability. Rather than relying on complex network redesign, the proposed framework offers a practical and effective pipeline for nighttime image dehazing.

Figures & tables

Figure 1: Pipeline overall pipeline of our proposed solution, showing CLIP-guided data augmentation, stage-wise NAFNet training, and inference-time enhancement.

Figure 2: Qualitative comparison on representative NHM-20 samples. From left to right: Input, NAFNet-100k, Weighted Ensemble, and ground truth.

Data Source	# Samples	Description
NTHazy	25	Target-domain paired data
I-HAZE (selected)	21	External supplementary data
Dense-Haze (selected)	10	External supplementary data
HAZE1K (selected)	3	External supplementary data
Total	59	Selected training set size

Table 1: Composition of the training data.

Item	Setting
Basic network	NAFNetLocal + NAFNet
Public inference structure	Width = 32, encoder blocks $[1,1,1,28]$ , one middle block, decoder blocks $[1,1,1,1]$
Input preprocessing	Crop to multiples of 8 before inference; NHM public reproducible evaluation inputs are uniformly saved as PNG
Stage I data	NTHazy (25 pairs)
Stage II data	NTHazy + filtered I-HAZE / Dense-Haze / HAZE1K (59 pairs in total)
Training configuration	Batch size = 4, patch size = $256\times 256$ , 200K iterations per stage

Table 2: Summary of the implementation and inference settings.

Method	PSNR-Y $\uparrow$	SSIM-Y $\uparrow$	PSNR-RGB $\uparrow$	SSIM-RGB $\uparrow$	LPIPS-Alex $\downarrow$
Input	23.4697	0.887581	19.4925	0.811362	0.278838
NAFNet-80k	24.6568	0.906439	18.2269	0.776861	0.310485
NAFNet-100k	25.0125	0.904238	18.7441	0.782752	0.311571
NAFNet-200k	24.8830	0.904796	18.3753	0.763607	0.318563
Weighted Ensemble	24.8875	0.905079	18.3813	0.764708	0.317944

Table 3: Results on the NHM-20 public reproducible evaluation.

Image Dehazing

Image dehazing restores clear scene details and colors from images obscured by haze, fog, or smoke.

See all 35 papers

HistoFusionNet: Histogram-Guided Fusion and Frequency-Adaptive Refinement for Nighttime Image Dehazing
Apr 4, 2026Mohammad Heydari, Wei Dong, Shahram Shirani +2Image Dehazing Image Restoration
McMaster University
Bilevel Layer-Positioning LoRA for Real Image Dehazing
Mar 11, 2026Yan Zhang, Long Ma, Yuxin Feng +3Image Dehazing Low-Rank Adaptation
School of Computer Science and Engineering, Sun Yat-sen University · Dalian University of Technology · Xidian University +1
HazeSpikeMamba: Coupling Spiking-Inspired and State-Space Features for Self-Supervised Real-World Dehazing
Aug 7, 2026Haoran Liu, Huibin Li, Mingzhe Liu +2Unsupervised Domain Adaptation Spiking Neural Networks
School of Artificial Intelligence and Electronic Engineering, Sichuan Technology and Business University, Chengdu 611745, China · Chengdu University of Technology, College of Nuclear Technology and Automation Engineering, Chengdu 610059, China · Sigray, Inc., Concord, CA 94520, USA

Apr 4, 2026J/K move · Enter open · S save

HistoFusionNet: Histogram-Guided Fusion and Frequency-Adaptive Refinement for Nighttime Image Dehazing

Mohammad Heydari, Wei Dong, Shahram Shirani, Jun Chen+1

McMaster University

Image Dehazing Image Restoration

Open paper AskPDF arXiv

Nighttime image dehazing remains a challenging low-level vision problem due to the joint presence of haze, glow, non-uniform illumination, color distortion, and sensor noise, which often invalidate assumptions commonly used in daytime dehazing. To address these challenges, we propose HistoFusionNet, a transformer-enhanced architecture tailored for nighttime image dehazing by combining histogram-guided representation learning with frequency-adaptive feature refinement. Built upon a multi-scale encoder-decoder backbone, our method introduces histogram transformer blocks that model long-range dependencies by grouping features according to their dynamic-range characteristics, enabling more effective aggregation of similarly degraded regions under complex nighttime lighting. To further improve restoration fidelity, we incorporate a frequency-aware refinement branch that adaptively exploits complementary low- and high-frequency cues, helping recover scene structures, suppress artifacts, and enhance local details. This design yields a unified framework that is particularly well suited to the heterogeneous degradations encountered in real nighttime hazy scenes. Extensive experiments and highly competitive performance of our method on the NTIRE 2026 Nighttime Image Dehazing Challenge benchmark demonstrate the effectiveness of the proposed method. Our team ranked 1st among 22 participating teams, highlighting the robustness and competitive performance of HistoFusionNet. The code is available at: https://github.com/heydarimo/Night-Time-Dehazing

Abstract

Image Dehazing

Explore similar work

HistoFusionNet: Histogram-Guided Fusion and Frequency-Adaptive Refinement for Nighttime Image Dehazing

Bilevel Layer-Positioning LoRA for Real Image Dehazing

HazeSpikeMamba: Coupling Spiking-Inspired and State-Space Features for Self-Supervised Real-World Dehazing