• 📄 Thrilled to share our work was Accepted by NeurIPS-25!
• 📜 Our paper is available at Arxiv!

Low-light vision remains a fundamental challenge in computer vision due to severe illumination degradation, which significantly affects the performance of downstream tasks such as detection and segmentation. While recent state-of-the-art methods have improved performance through invariant feature learning modules, they still fall short due to incomplete modeling of low-light conditions. Therefore, we revisit low-light image formation and extend the classical Lambertian model to better characterize low-light conditions. By shifting our analysis to the frequency domain, we theoretically prove that the frequency-domain channel ratio can be leveraged to extract illumination-invariant features via a structured filtering process. We then propose a novel and end-to-end trainable module named Frequency-domain Radial Basis Network (FRBNet), which integrates the frequency-domain channel ratio operation with a learnable frequency domain filter for the overall illumination-invariant feature enhancement. As a plug-and-play module, FRBNet can be integrated into existing networks for low-light downstream tasks without modifying loss functions. Extensive experiments across various downstream tasks demonstrate that FRBNet achieves superior performance, including +2.2 mAP for dark object detection and +2.9 mIoU for nighttime segmentation.

Python 3.7+, CUDA 9.2+, Pytorch 1.8+
Our implementation is based on the latest MMDetection 3.x version. For more detailed information, please refer to here.

conda creat --name FRBNet python=3.8 -y
conda activate FRBNet
conda install pytorch torchvision -c pytorch
pip install -U openmim
mim install mmengine
mim install mmcv

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -v -e .

1. Put the custom_mmlab/FRBNet_mmdet/configs inside the mmdetection/configs
2. Put custom_mmlab/FRBNet_mmdet/mmdet/datasets/exdark_voc.py and custom_mmlab/FRBNet_mmdet/mmdet/datasets/dark_face.py inside mmdetection/mmdet/datasets/
3. Need to update the mmdetection/mmdet/datasets/__ init __.py as

from .exdark_voc import ExDarkVocDataset
from .dark_face import DarkFaceDataset

__all__ = [
    ......, 'ExDarkVocDataset', 'DarkFaceDataset'
]

4. Put the custom_mmlab/FRBNet_mmdet/mmdet/mdoels/detectors/frbnet_utils.py and custom_mmlab/FRBNet_mmdet/mmdet/mdoels/detectors/frbnet.py inside the mmdetection/mmdet/mdoels/detectors/
5. Need to update the mdetection/mmdet/mdoels/detectors/__ init __.py as

from .frbnet import FRBNet
__all__ = [
    ......, 'FRBNet'
]

The pretrained model on the COCO dataset can be download here.

To facilitate reproduction, we will also provide a version without MMDet in the near future. You can easy to start FRBNet directly!

ExDark You can download the official ExDark dataset from this link. The expected directory structure is as follows:

dataset
|-ExDark
| |--Annotations
|  |----2015_00001.png.xml
|  |----2015_00002.png.xml
|  |----2015_00003.png.xml
|  |----2015_0.....png.xml
|  |----...
| |--JPEGImages
|  |----2015_00001.png
|  |----2015_00002.png
|  |----2015_00003.png
|  |----2015_0.....png
|  |----...
| |--train.txt
| |--test.txt
| |--val.txt

More details can be found at here
DarkFace You can download the official DarkFace dataset from this link. The expected directory structure is as follows:

dataset
|-ExDark
| |--Annotations
|  |----1.xml
|  |----2.xml
|  |----3.xml
|  |----4.xml
|  |----...
| |--JPEGImages
|  |----...
| |--train.txt
| |--test.txt
| |--val.txt

ACDC-Night You can download the official ACDC dataset from this link. The expected directory structure is as follows:

dataset
|-ACDC
| |--new_gt_labelTrainIds
|  |----train
|    |------...
|  |----val
|    |------...
| |--rgb
|  |----train
|    |------...
|  |----val
|    |------...
|  |----test
|    |------...

LIS You can download the official LIS dataset from this link. The expected directory structure is as follows:

dataset
|-LIS
| |--annotations
|  |----lis_coco_JPG_train+1.json
|  |----lis_coco_JPG_test+1.json
|  |----...
| |--RGB-dark
|  |----JPEGImages
|    |------2.JPG
|    |------4.JPG
|    |------6.JPG
|    |------...
| |--RAW-dark
|  |----JPEGImages
|    |------2.png
|    |------4.png
|    |------6.png
|    |------...

⭐ Note: We will provide complete dataset packages via Google Drive or Baidu Netdisk in the future.

Single GPU

python tools/train.py configs/yolov3_frbnet_exdark.py

Multiple GPU

./tools/dist_train.sh configs/yolov3_frbnet_exdark.py GPU_NUM

Single GPU

python tools/test.py configs/yolov3_frbnet_exdark.py [model path] --out [NAME].pkl

Multiple GPU

./tools/dist_train.sh configs/yolov3_frbnet_exdark.py [model path] GPU_NUM --out [NAME].pkl

We provide a model weights based on YOLOv3 for low-light object detection on the ExDark dataset. You can download the .pth from here (organizing) as [model path] to test.

In this project we use parts of the official implementations of the following works:

• MMdetection: mmdetection
• MMsegmentation: mmsegmentation
• YOLA: You Only Look Around
• FeatEnHancer: FeatEnHancer
• MAET: Multitask AET

We gratefully acknowledge the authors who have open-sourced their methods. In the same spirit, we are committed to releasing the complete version of our code in the near future.