Exposure Bracketing Is All You Need For A High-Quality Image
Zhilu Zhang, Shuohao Zhang, Renlong Wu, Zifei Yan, Wangmeng Zuo

TL;DR
This paper introduces a novel approach using exposure bracketing and a temporally modulated recurrent network to enhance low-light image quality by combining multiple images, with a focus on synthetic pre-training and real-world adaptation.
Contribution
It proposes a new method leveraging exposure bracketing with a TMRNet and self-supervised adaptation, addressing the challenge of real-world data collection.
Findings
Outperforms state-of-the-art multi-image processing methods
Effective in real-world nighttime scenarios
Demonstrates the benefit of synthetic pre-training and adaptation
Abstract
It is highly desired but challenging to acquire high-quality photos with clear content in low-light environments. Although multi-image processing methods (using burst, dual-exposure, or multi-exposure images) have made significant progress in addressing this issue, they typically focus on specific restoration or enhancement problems, and do not fully explore the potential of utilizing multiple images. Motivated by the fact that multi-exposure images are complementary in denoising, deblurring, high dynamic range imaging, and super-resolution, we propose to utilize exposure bracketing photography to get a high-quality image by combining these tasks in this work. Due to the difficulty in collecting real-world pairs, we suggest a solution that first pre-trains the model with synthetic paired data and then adapts it to real-world unlabeled images. In particular, a temporally modulated…
Peer Reviews
Decision·ICLR 2025 Poster
1. A novel setting combing several image restoration tasks, utilizing raw space images. 2. The performance is good enough. 3. Extensive experiments are done to prove characteristics of the proposed method.
1. The BracketIRE and BracketIRE+ are similar with all-in-one. The authors should discuss on this issue. 2. The computational cost seems to be high. The fairness of comparison is doubtful. 3. The datasets must be publicly, otherwise the work is not so essential.
1. This work is well-complete overall, including data simulation, real-world data collection, and method design and experiments. 2. The experimental results show improvements in both visual quality and quantitative metrics, demonstrating its effectiveness.
1. The comparison methods in Table 2 are all trained on synthetic data, while TMRNet is fine-tuned on real-world data. This comparison is somewhat unfair. It is recommended that the authors include the results of TMRNet without fine-tuning on real-world data and provide a note. 2. This paper claims that more frames result in better performance, would it introduce frames with over-exposure or large blur, making the restoration process more challenging? 3. In summarizing the contributions, the aut
1. The proposed method is the first method that performs the four restoration and enhancement tasks altogether. 2. The necessity of the temporal modulated recurrent network is clearly explained, and its effectiveness is supported by experimental results. 3. The necessity of the introduced loss functions is clearly explained, and their effectiveness is supported by experimental results. 4. The constructed synthetic and real datasets will be helpful for researchers in the related field. 5. The
1. I understand that ICLR accepts application-oriented papers. However, from the theoretical point-of-view, the manuscript does not contain sufficient technical novelties. The network can perform the four different tasks altogether since it is simply trained using the synthetic (and real) dataset containing four degradation. In other words, except for the dataset, the network does not have significant novelties compared to existing video restoration networks. 2. The proposed temporal modulati
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMedical Imaging Techniques and Applications · Photoacoustic and Ultrasonic Imaging · Radiomics and Machine Learning in Medical Imaging
MethodsFocus
