DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment

Wuqi Wang; Haochen Yang; Baolu Li; Jiaqi Sun; Xiangmo Zhao; Zhigang Xu; Qing Guo; Haigen Min

cs.CVMar 18, 2026

DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment

Authors: Wuqi Wang, Haochen Yang, Baolu Li, Jiaqi Sun, Xiangmo Zhao, Zhigang Xu, Qing Guo, Haigen Min, +2 more

Organizations: Chang’an University. · Cleveland State University. · A*STAR.

Abstract

The low-light conditions are challenging to the vision-centric perception systems for autonomous driving in the dark environment. In this paper, we propose a new benchmark dataset (named DarkDriving) to investigate the low-light enhancement for autonomous driving. The existing real-world low-light enhancement benchmark datasets can be collected by controlling various exposures only in small-ranges and static scenes. The dark images of the current nighttime driving datasets do not have the precisely aligned daytime counterparts. The extreme difficulty to collect a real-world day and night aligned dataset in the dynamic driving scenes significantly limited the research in this area. With a proposed automatic day-night Trajectory Tracking based Pose Matching (TTPM) method in a large real-world closed driving test field (area: 69 acres), we collected the first real-world day and night aligned dataset for autonomous driving in the dark environment. The DarkDriving dataset has 9,538 day and night image pairs precisely aligned in location and spatial contents, whose alignment error is in just several centimeters. For each pair, we also manually label the object 2D bounding boxes. DarkDriving introduces four perception related tasks, including low-light enhancement, generalized low-light enhancement, and low-light enhancement for 2D detection and 3D detection of autonomous driving in the dark environment. The experimental results show that our DarkDriving dataset provides a comprehensive benchmark for evaluating low-light enhancement for autonomous driving and it can also be generalized to enhance dark images and promote detection in some other low-light driving environment, such as nuScenes.The code and dataset will be publicly available at https://github.com/DriveMindLab/DarkDriving-ICRA-2026.

Figures & tables

Panel of Figure 1 — Fig. 1 : Data examples of our DarkDriving and other representative low-light enhancement and nighttime driving datasets : (a) our DarkDriving with precisely aligned location and spatial contents (error: centimeters) of day and night, (b) low-light enhancement dataset by controlling exposures, and (c) nighttime driving dataset with roughly aligned daytime location (error: meters) and unaligned spatial contents using the GPS signal.

Dataset	Year	Real/ Simulated	Independent of Exposure Control	RGB Images	Scene	Day and Night Paired Images	Pair Alignment Error
LOL [ 1 ]	2018	Real	$\times$	1,000	Static	✓	-
SCIE [ 2 ]	2018	Real	$\times$	4,413	Static	✓	-
SID [ 3 ]	2018	Real	$\times$	5,094	Static	✓	-
BBD100K-dark [ 4 ]	2020	Real	✓	14K	Driving	$\times$	-
NuScenes-night [ 5 ]	2020	Real	✓	4.6K	Driving	$\times$	-
Dark Zurich [ 6 ]	2020	Real	✓	8,779	Driving	✓	Meters

TABLE I : Comparison of our DarkDriving dataset and other low-light enhancement and nighttime driving datasets.

Fig. 2 : Setup of sensors on the automated vehicle and their coordinate systems.

Sensors	Details
1×Camera	RGB, 2,448×2,048 resolution, 41° FOV
1×LiDAR	128 channels, 200m capturing range, -25° to +15° vertical FOV, Range Precision: ±2cm error
GPS & IMU	Xsens MTi-680G, 400 Hz SDI

TABLE II : Specification of sensors on automated vehicle.

Fig. 3 : Details of the large real-world closed driving test field at Chang’an University. From top to bottom: satellite image of driving test field, local and global demos of its High-Precision Point Cloud Map.

Fig. 4 : The pipeline of the automatic day-and-night Trajectory Tracking based Pose Matching (TTPM) method for data collection. One example of the desired trajectories is shown here. Best view in color.

Panel of Figure 7 — Fig. 5 : Dark driving scenarios of our DarkDriving dataset. One sample day-night image pair is shown for each dark driving scenario, followed by its percentage distribution in our dataset.

	Full-reference Metrics			Non-reference Metrics
Methods	PSNR↑	SSIM↑	LPIPS↓	MUSIQ↑	NIQE↓	HyperIQA↑	CNNIQA↑
Night Image	6.58	0.38	0.59	47.79	6.58	0.34	0.54
LLFormer	22.95	0.79	0.41	26.81	6.28	0.25	0.27
SNR-SKF	20.29	0.74	0.39	35.80	5.86	0.32	0.40
Retinexformer	23.66	0.80	0.34	33.18	6.06	0.31	0.34
ControlNet	13.17	0.58	0.50	65.16	5.32	0.66	0.67

TABLE III : Quantitative comparison of image quality for SOTA low-light enhancement methods on the testing set of DarkDriving . The best and second best results are highlighted in bold and underline , respectively.

Panel of Figure 9 — Fig. 6 : Visual comparison of nighttime image enhancement methods on the testing set of DarkDriving. The input is the nighttime image (a) and the outputs (b–f) are to approximate the corresponding daytime image (g).

	Non-reference Metrics
Pre-trained Dataset	MUSIQ↑	NIQE↓	HyperIQA↑	CNNIQA↑
LOL-v2-real	28.23	4.63	0.30	0.51
LOL-v2-synthetic	28.45	4.44	0.27	0.54
DarkDriving	35.83	4.41	0.31	0.53

TABLE IV : Cross-dataset Enhancement. Pre-trained on LOL-v2 or DarkDriving using SNR-Aware [ 18 ] , Tested on the nighttime validation set of nuScenes.

Panel of Figure 11 — Fig. 7 : Visual Cross-dataset Enhancement on nuScenes .

Method	YOLOv11-pretrained		YOLOv11-finetune
Method	AP 50 ↑	AP 50-90 ↑	AP 50 ↑	AP 50-90 ↑
Night	0.316	0.178	0.166	0.083
CLIP-LIT ∗ [ 32 ]	0.309	0.174	0.183	0.093
EnlightenGAN ∗ [ 33 ]	0.291	0.165	0.093	0.048
LLFormer [ 20 ]	0.161	0.090	0.167	0.071
Retinexformer [ 21 ]	0.173	0.090	0.215	0.105

TABLE V : 2D Detection on Darkdriving nighttime test set . ∗ : unsupervised enhancement approaches. Other enhancement methods: supervised and trained on the DarkDriving training set.

Panel of Figure 13 — Fig. 8 : 2D detection errors on nighttime images can be reduced after low-light enhancement. (a) and (d) are night-day paired data of DarkDriving dataset.

	BEVDepth (RGB)			CRN (RGB+Radar)
Method	AP↑	ATE↓	AOE↓	AP↑	ATE↓	AOE↓
Night	0.134	0.787	0.957	0.435	0.474	0.597
CLIP-LIT ∗ [ 32 ]	0.131	0.791	0.972	0.430	0.475	0.607
EnlightenGAN ∗ [ 33 ]	0.138	0.786	0.948	0.436	0.473	0.585
LightTheNight [ 11 ]	0.176	0.774	1.108	0.485	0.478	0.464
SNR-Aware [ 18 ]	0.148	0.786	0.911	0.481	0.468	0.575

TABLE VI : 3D Detection on nuScenes nighttime validation set . BEVDepth (RGB-Only) [ 34 ] and CRN (RGB+Radar) [ 35 ] are trained on nuScenes daytime training set. ∗ : unsupervised enhancement approaches. LightTheNight: trained on nuScenes. SNR-Aware: pre-trained on DarkDriving training set (same as that in Table IV ).

Explore similar work

Oct 14, 2026

OLiVES: An Outdoor Low-Light Video Benchmark for Enhancement and Segmentation

Low-light video processing remains an underexplored problem in computer vision, primarily due to the scarcity of high-quality ground-truth data. Existing public datasets exhibit several limitations, including restricted motion diversity, limited scene coverage, weak degradations, reliance on auxiliary modalities such as event data, and misalignment between low-light inputs and ground truth. Furthermore, current datasets typically address either low-light enhancement or scene understanding, whereas these tasks should be considered jointly. We introduce OLiVES, a benchmark dataset for outdoor low-light videos with dynamic scenes, designed to support both low-light video enhancement (LLVE) and video object segmentation (VOS). The dataset addresses key content gaps by incorporating a diverse range of natural textures, animals, occlusions, dynamic activities, and close-up scenes. To the best of our knowledge, OLiVES is the first outdoor low-light video dataset captured using professional cinema-grade cameras. It provides over 25,000 aligned normal- and low-light frame pairs, along with over 46,000 VOS annotations. We benchmark state-of-the-art LLVE and VOS methods on OLiVES and show that training on it improves performance on both tasks. The OLiVES project page is available at: https://anonymous-OLiVES.github.io/.

Yini Li, Joanne Lin, David Blackham +2

Visual Information Laboratory, University of Bristol, Bristol, United Kingdom

Low-Light Image Enhancement Video Object Segmentation

Ask PDF

Dec 9, 2025

LiDAS: Lighting-driven Dynamic Active Sensing for Nighttime Perception

Nighttime environments pose significant challenges for camera-based perception, as existing methods passively rely on the scene lighting. We introduce Lighting-driven Dynamic Active Sensing (LiDAS), a closed-loop active illumination system that combines off-the-shelf visual perception models with high-definition headlights. Rather than uniformly brightening the scene, LiDAS dynamically predicts an optimal illumination field that maximizes downstream perception performance, i.e., decreasing light on empty areas to reallocate it on object regions. LiDAS enables zero-shot nighttime generalization of daytime-trained models through adaptive illumination control. Trained on synthetic data and deployed zero-shot in real-world closed-loop driving scenarios, LiDAS enables +18.7% mAP50 and +5.0% mIoU over standard low-beam at equal power. It maintains performances while reducing energy use by 40%. LiDAS complements domain-generalization methods, further strengthening robustness without retraining. By turning readily available headlights into active vision actuators, LiDAS offers a cost-effective solution to robust nighttime perception.

Simon de Moreau, Andrei Bursuc, Hafid El-Idrissi +1

Valeo - Mines Paris PSL · Valeo · Mines Paris PSL

Autonomous Driving Perception Active Sensing Active Visual Perception

Ask PDF

Apr 26, 2026

Reading in the Dark: Low-light Scene Text Recognition

Accurate text recognition in low-light environments is essential for intelligent systems in applications ranging from autonomous vehicles to smart surveillance. However, challenges such as poor illumination and noise interference remain underexplored. To address this gap, we introduce LSTR, a large-scale Low-light Scene Text Recognition dataset comprising 11,273 low-light images generated from well-lit datasets (ICDAR2015, IIIT5K, and WordArt), along with ESTR, which includes 60 real nighttime street-scene images in English and Spanish for exclusive evaluation. We explore two solution strategies: (1) employing Optical Character Recognition (OCR) models with fine-tuning and LoRA-based fine-tuning and (2) a joint training strategy that integrates a low-light image enhancement (LLIE) module with an OCR model. In particular, we propose a novel re-render LLIE (RLLIE) module, which demonstrates improved performance on real-world data. Through extensive experimentation, we analyze various training strategies and address a key research question: How bright is bright enough for effective scene text recognition? Our results indicate that standalone LLIE or OCR models perform inadequately under low-light conditions, highlighting the advantages of specialized, jointly trained text-centric approaches. Additionally, we provide a comprehensive benchmark to support future research in robust low-light scene text recognition. https://huggingface.co/datasets/lumimusta/Low-light_Scene_Text_Dataset.

Xuanshuo Fu, Lei Kang, Ernest Valveny +2

Computer Vision Center, Barcelona, Spain

Scene Text Recognition Low-Light Image Enhancement Optical Character Recognition

Ask PDF