File size: 2,466 Bytes
4750c51
 
 
 
5cda33a
 
 
 
b1e6c35
ef780ef
e69fdf1
203527c
80350c7
d4059ef
ef780ef
 
 
 
 
 
31d534c
ef780ef
d4059ef
aad7ce9
 
 
 
 
 
ef780ef
d4059ef
80350c7
b1e6c35
 
d4059ef
b1e6c35
 
e9abbf3
832f724
b1e6c35
1253b13
832f724
 
e9abbf3
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
---
license: bsd-3-clause
---

<h1>
Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration	
</h1>

<a rel="AAA" href="https://arxiv.org/pdf/2505.16479">Paper</a>  | 
<a rel="AAA" href="https://github.com/henlyta/ClearNight">Github</a>  | 
<a rel="AAA" href="https://henlyta.github.io/ClearNight/index.html">Page</a>  | 
<a rel="AAA" href="https://huggingface.co/datasets/YuetongLiu/AllWeatherNight">Dataset</a>

<h2>AllWeatherNight</h2>

We observe that uneven lighting conditions in real-world nighttime scenes often interact with weather degradations. 
To synthesize more realistic nighttime images with adverse weather conditions, we introduce an illumination-aware degradation generation approach. 
We show four different synthetic image variants of nighttime scenes. 
Weather Only and Flare Only denote synthesis with illumination-aware weather degradation and flare, respectively. Ours involves synthesis with both types of degradations.

<img src="https://github.com/henlyta/ClearNight/blob/page/static/image/dataset2.png?raw=True">

<h2>Dataset Statistics</h2>

We synthesize 8,000 nighttime images for model training, encompassing both multi-degradation and single-degradation scenarios with various degradation scales, directions, patterns and intensities. 
The test dataset consists of two parts: a synthetic subset and a real subset, each containing 1,000 images.
The synthetic subset evaluates models across 7 dimensions, covering synthetic images with both multiple and single degradations.
The 1,000 collected real-world images are categorized into 4 different degradation types and serve as the real subset for assessing models in real-world scenarios.


<h2>Intended Use</h2>

Our AllWeatherNight dataset is released under the BSD 3-Clause License, a permissive open-source license that grants users the freedom to use, copy, modify, and distribute the dataset, whether in its original form or as part of derivative works. The license is employed on generated degraded images and labels. The ground-truths from BDD100K and Exdark adhere to the BSD 3-Clause License.

<h2>Citation</h2>

If you find our work is helpful to your research, please cite the papers as follows:
<pre>
@inproceedings{aaai2026clearnight,
      title={Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration}, 
      author={Liu, Yuetong and Xu, Yunqiu and Wei, Yang and Bi, Xiuli and Xiao, Bin},
      year={2026},
      booktitle={AAAI}
}
</pre>