Efficient RAW Image Deblurring with Adaptive Frequency Modulation

Wenlong Jiao, Binglong Li, Wei Shang, Ping Wang, Dongwei Ren

---

Abstract: Image deblurring plays a crucial role in enhancing visual clarity across various applications. Although most deep learning approaches primarily focus on sRGB images, which inherently lose critical information during the image signal processing pipeline, RAW images, being unprocessed and linear, possess superior restoration potential but remain underexplored. Deblurring RAW images presents unique challenges, particularly in handling frequency-dependent blur while maintaining computational efficiency. To address these issues, we propose Frequency Enhanced Network (FrENet), a framework specifically designed for RAW-to-RAW deblurring that operates directly in the frequency domain. We introduce a novel Adaptive Frequency Positional Modulation module, which dynamically adjusts frequency components according to their spectral positions, thereby enabling precise control over the deblurring process. Additionally, frequency domain skip connections are adopted to further preserve high-frequency details. Experimental results demonstrate that FrENet surpasses state-of-the-art deblurring methods in RAW image deblurring, achieving significantly better restoration quality while maintaining high efficiency in terms of reduced MACs. Furthermore, FrENet's adaptability enables it to be extended to sRGB images, where it delivers comparable or superior performance compared to methods specifically designed for sRGB data. The source code will be publicly available.

---

Network Architecture

QuickRun

Install

conda create -n your_env python=3.9
conda activate your_env
pip install -r requirements.txt

Training

1. Download Deblur-RAW training and testing data.
2. Please change the dataset path in options/frenet.yml to the path of your dataset and run

torchrun --nproc_per_node=2 train.py -config options/frenet.yml 

Evaluation

Please download the pre-trained deblurring model to the checkpoints folder and run

python eval_raw.py -config options/frenet.yml

Generating DNG with Original Metadata

To ensure the deblurred RAW output can be correctly processed by standard RAW editors, it's crucial to embed the original camera's metadata into the final DNG file. We provide scripts to facilitate this process. The method involves first extracting metadata from the blurry RAW file and then injecting it into the deblurred output.

This process requires ExifTool. Please install it first.

Use metadata.py to extract the metadata from the original blurry RAW image. After processing the blurry RAW image through the model, use add_exif.py to inject the extracted metadata into the deblurred result. Afterward, you can use eval_rgb_from_raw.py to test the model's performance in the RGB domain following this ISP pipeline.

Results

FreNet achieved state-of-the-art (SOTA) performance in RAW image deblurring.

Citation

@misc{jiao2025efficient,
    title={Efficient RAW Image Deblurring with Adaptive Frequency Modulation},
    author={Wenlong Jiao and Binglong Li and Wei Shang and Ping Wang and Dongwei Ren},
    year={2025},
    eprint={2505.24407},
    archivePrefix={arXiv},
    primaryClass={eess.IV}
}

Contact

If you have any question, please contact [email protected] or [email protected]

Acknowledgment

Some codes are based on NAFNet and BasicSR toolbox. Thanks for their awesome works.