Dear Reviewer, vrds:

Thank you for your valuable suggestions. We have carefully revised the paper to address your concerns to the fullest extent possible. Specifically, we have added Section A and Section B in the supplementary materials (pages 16–20) to provide detailed explanations about the imaging process, underlying principles, and dataset scale. These additions aim to clarify and enhance the understanding of the key aspects you highlighted.

Below are specific answers to your questions.

Q.1 Concern about Dataset Size for ISP tasks:

Thank you for your suggestion and concern. We have answered the question about the scale of the dataset in Summary of Official Reviews (2/2). We hope that this will resolve your doubts.

Q.2 Open-Sourcing the Dataset

We promise that the dataset, benchmark code and pre-train model will be open sourced upon the acceptance of this paper. This will allow the broader community to access, evaluate, and build upon our work. A clear statement regarding dataset release has been added to the paper.

Q.3 Concern about Reproducibility

Thank you for your concern regarding the reproducibility of our experiments with the sensor. We appreciate this opportunity to provide further clarification. We would like to emphasize that the sensor is expected to be commercialized and made publicly available in the near future.

Our research serves as an early-stage exploration of its potential applications, aiming to uncover the benefits and challenges of integrating such technology into ISP tasks. We hope that this work can inspire further advancements in this field and help guide the community's adoption and innovation around this promising technology.

Q.4 & Q.5 More Background of Dataset & More Explain about Events

Thank you for your suggestion. We agree that clarifying the dataset’s background and unique characteristics will greatly benefit readers. In the appendix, we have added section A to explain the imaging principles, features, and unique advantages of this sensor. Additionally, we have included new examples to highlight its performance in challenging scenarios, such as high-speed motion and low-light conditions, demonstrating the benefits of this hybrid sensor.

Best

Thank you for your thoughtful review. In the revision paper, we have incorporated additional background explanations to address your concerns. We hope these revisions provide clarity and resolve the issues you raised. We look forward to engaging in further discussions with you.

Dear Reviewer fVak:

Thank you for your thoughtful feedback and valuable suggestions. We greatly appreciate the time and effort you have taken to review our work. Below, we provide detailed responses to your concerns and describe how we have addressed them in the revised manuscript.

Q.1 Concern about Dataset Size

Thank you for pointing this out. This concern has been discussed in detail in the section "Summary and Answers of Official Reviews (2/2)." Please refer to this section for our comprehensive response.

Q.2 Elaboration on the Role of Event Data in Each ISP Stage

Thank you for raising this important point. To clarify, in the Section CONTROLLABLE ISP stage, we used a ColorChecker-based ISP pipeline to generate reference images as ground truth. In this stage, tasks such as demosaicing and color correction are performed using traditional computational methods. The evaluarion results are demonstrated in Figure 5 of the main paper. This process does not involve the use of event data.

Inspired by your suggestion, we have included additional analysis in the supplementary materials to explore how event characteristics could assist in improving RAW-level ISP tasks when integrated into deep learning frameworks. We believe this will provide valuable insights for future research directions and further demonstrate the potential of event-guided ISP.

Q.3 Refine Writing

Thank you for highlighting the need for improved clarity in writing. We have thoroughly reviewed the manuscript to correct spelling errors and ensure precise language throughout. Your feedback has helped us significantly improve the readability and robustness of the paper.

Q.4 Hyperparameters of Comparative Algorithms

Thank you for the suggestion. We have conducted additional experiments with CameraNet using optimized hyperparameters. These results, included in the revised manuscript, provide a more comprehensive and fair comparison of baseline methods.

Q.5 Alignment of EVS and APS

We appreciate your question regarding the alignment between EVS (Event-based Vision Sensor) and APS (Active Pixel Sensor) data. In the appendix (Section A, Pages 16–19), we provide a detailed explanation of the temporal alignment process. Specifically, the EVS and APS sensors are equipped with unified timestamps, enabling precise alignment at the pixel level. This ensures that event data is synchronized with the rolling shutter frames of the APS output, allowing for accurate integration in hybrid sensor tasks.

Once again, we thank you for your constructive feedback. Your comments have not only helped us address key concerns but also inspired us to refine our work and present a more robust contribution to the field. We hope our responses and revisions address your concerns satisfactorily.

Best

Thanks for your time and attention. In the revision paper, we have addressed the dataset scale, the data alignment methodology, and provided additional background information. We hope you have had the opportunity to review the revised version. We are eager to discuss any further questions or feedback you may have.

Dear Reviewer qDHH,

Thank you for your thoughtful review and for highlighting the key contributions of our work. We greatly appreciate your recognition of the following strengths. (1) Dataset: The first of its kind with aligned RAW and event data from a hybrid vision sensor, providing new opportunities for ISP algorithm development. (2) Benchmark Task: Evaluating the dataset with various ISP methods and deriving valuable insights for event-guided ISP research. Below, we address each of your concerns in detail.

Q.1 Dataset Scale and Diversity

Thank you for raising this important point. While our dataset consists of 3,373 high-resolution, real-world images, making it the larger datasets for hybrid sensor ISP tasks now. For a more detailed response, please refer to "Summary and Answers of Official Reviews (2/2)."

Additionally, we plan to maintain and expand this dataset as a dynamic, growing resource for the research community. We aim to enhance its diversity while ensuring its focus on high-quality, real-world data.

Q.2 Reproducibility and Dataset Usage

We greatly appreciate your suggestion regarding early release. However, to comply with double-blind review policies, we are unable to release the dataset during the review process. We are fully committed to making the dataset, benchmark code, and results publicly available immediately upon acceptance of the paper. This timeline ensures fairness and accessibility for the research community.

Q.3 Event-Guided ISP Baseline Evaluation

Thank you for this constructive suggestion. As the first dataset specifically designed for event-guided ISP tasks, our primary goal was to establish a baseline by demonstrating the utility of a simple event fusion approach. Prior to this work, no research has explored using events to guide the RAW ISP process.

We recognize the importance of further developing and evaluating more event-guided architectures. To address this, we have included discussions in the revised manuscript to emphasize future directions, including advanced event-guided ISP methods and architectures.

Q.4 Dataset Scale and Diversity:

Thank you for this question. As mentioned, our dataset is already larger than other similar datasets and sufficient for training and testing ISP models effectively.

We are also committed to expanding its scale and diversity over time by incorporating additional data, scenes, and lighting conditions. This dynamic approach will address current limitations and better serve the research community.

Q.5 Early Data Release:

Thank you for raising this question. To ensure accessibility and reproducibility, we will release the dataset, benchmark codes, and experimental results immediately upon acceptance of the paper. This timeline ensures compliance with double-blind review policies while supporting the broader research community.

Q.6 Event-Guided ISP Baseline Evaluation:

As this is the first dataset designed for event-guided ISP tasks, we demonstrated the potential of a simple baseline approach in this work.

Once again, we sincerely thank you for your constructive feedback. Your comments have been invaluable in helping us improve the paper and refine its contributions. We hope our responses address your concerns and demonstrate the robustness of our work.

Sincerely,

Thank you for your time and attention. In the revision paper, we have elaborated on the dataset scale and provided additional examples to showcase its diversity. Additionally, we have committed to releasing the dataset and benchmark code upon acceptance of the paper. We hope you have had the opportunity to review our revisions and look forward to further engaging discussions with you.

Dear Reviewer AvAS,

Thank you for taking the time to review our work and for providing thoughtful and constructive feedback. We are grateful for your recognition of the following strengths in our paper: (1) Background: Clearly explained and accessible to readers. (2) Relevance: RGB-event ISP is an interesting and meaningful topic for future cameras. (3) Writing: Well-structured and easy to understand. We have carefully reviewed your comments and suggestions. Below, we provide detailed responses to your specific concerns.

Q.1 Dataset Scale and Simulated Data Generation

Thank you for raising this concern. Compared to the most recent similar datasets, our dataset is four times larger in scale. More importantly, our dataset is based on real-world data and includes event outputs, which are critical for hybrid sensor ISP tasks. For a more detailed response, please refer to "Summary and Answers of Official Reviews (2/2)."

Additionally, we appreciate your suggestion about expanding the dataset. We are committed to the long-term maintenance and future expansion of this dataset to include more scenes and conditions. We also believe the current scale is already sufficient to support training and testing models effectively.

Q.2 Trustworthiness of Ground Truth Generated by MATLAB

We understand your concerns regarding the use of MATLAB-based tools for generating ground truth. To clarify:

• The MATLAB-based controllable ISP framework is a widely recognized tool in academic research for generating high-quality ground truth images. This framework allows us to systematically control key ISP stages, such as demosaicing [a], denoise [b] and color correction [c]. In traditional ISP tasks, this approach is a standard and reliable method in the academic community [d].

• To ensure transparency, we have included detailed descriptions of the framework in the manuscript and have conducted quantitative evaluations to validate the accuracy and consistency of the generated ground truth images. These are illustrated in Figure 5 of the main paper and Figure 13 of the Appendix.

Additionally, in practical scenarios, consumer cameras imaging often lacks precise color references. By including a ColorChecker in every scene and using it to generate accurate color correction matrices, our approach ensures greater color accuracy than methods without such reference data.

Q.3 Integration of Events into Traditional ISP

Thank you for this valuable suggestion. We have expanded the discussion of the advantages of integrating events into traditional ISP tasks in the revised manuscript. Specifically:

Practical Examples: In Section A, Figure 10 of the supplementary materials, we present examples in low-light and fast-motion scenarios. These demonstrate how events provide high temporal resolution and wide dynamic range, highlighting their advantages over RGB-based approaches.

Theoretical Analysis: Based on your suggestion, we have added a new section in the supplementary materials analyzing why event data is effective for ISP tasks, including demosaicing and color correction. This analysis provides a theoretical foundation for integrating events into ISP.

Once again, thank you for your constructive feedback. Your comments have significantly improved the clarity and impact of our work. We hope our responses adequately address your concerns and demonstrate the robustness of our contributions.

Best

Reference:

• [a] Malvar, H.S., L. He, and R. Cutler, High quality linear interpolation for demosaicing of Bayer-patterned color images. ICASPP, Volume 34, Issue 11, pp. 2274-2282, May 2004.
• [b] Metzler, Christopher A., Arian Maleki, and Richard G. Baraniuk. "BM3D-AMP: A new image recovery algorithm based on BM3D denoising." 2015 IEEE international conference on image processing (ICIP). IEEE, 2015.
• [c] Westland, Stephen, Caterina Ripamonti, and Vien Cheung. Computational colour science using MATLAB. John Wiley & Sons, 2012.
• [d] Sumner, Rob. "Processing raw images in matlab." Department of Electrical Engineering, University of California Sata Cruz 2 (2014).

Thank you for your appreciation of our work. In the revision paper, we have discussed the dataset scale in detail and provided additional examples. Additionally, we have demonstrated the advantages of this camera in motion and low-light scenarios. We hope the revision paper has addressed your concerns and captured your interest. We look forward to engaging with you further.

Dear Reviewer q8r6,

Thank you for your insightful review and for highlighting the significant contributions of our work. We are especially grateful for your recognition of: (1) Dataset Contribution. (2)Thoroughly assessing trainable ISP methods on the dataset, showcasing its potential. Below, we address your concerns in detail, incorporating your valuable suggestions to improve the clarity and quality of our work.

Q.1. Insufficient Dataset Information

We appreciate your suggestion. To address this, we have:

• (1) Added a detailed description of the dataset in Appendix Section B, including information on the types of scenes, exposure modes, and camera settings used.
• (2) Included more visual samples to better illustrate the dataset's diversity and highlight specific scenarios of interest. We hope these updates will enhance readers' understanding of the dataset's content and significance.

Q.2. Statistical Dataset Information

Thanks for your suggestions. We have included a statistical summary table in the supplementary materials. This table provides a clear breakdown of the dataset by scene type, exposure mode, and camera-specific settings, offering a more comprehensive understanding of its structure and diversity.

Q.3.Performance Analysis of ISP Tasks

Thanks for your suggestions. We would like to clarify that the controllable ISP framework used in this work relies on ColorChecker-based evaluations to validate the accuracy of the generated RGB images, as shown in Figure 5. While intermediate processes such as demosaicing and white balancing are difficult to evaluate quantitatively due to the absence of ground truth, the final evaluation results provide an overall assessment of the pipeline's performance.

For the learning-based methods, we have included comprehensive evaluations using objective metrics (e.g., PSNR, SSIM, L1) as well as non-reference metrics (e.g., NIQE, PI), which collectively demonstrate the effectiveness of our dataset in supporting ISP research.

Q.4. Figures 6 and 7

Thank you for pointing this out. In the revised manuscript: Figures 6 and 7 have been combined to reduce redundancy. Indoor data visualization has been added to Figure 8, providing a more diverse set of examples.

Q.5. Comprehensive Dataset Performance Analysis

We have updated the supplementary materials to include a combined evaluation of indoor and outdoor data. This comprehensive analysis provides a more holistic view of the dataset’s overall performance, addressing your concern.

Q.6. Integrated Improvement Method

Thank you for highlighting this. The event-guided ISP baseline introduced in our work is intended as a simple starting point to demonstrate the potential of integrating event data into ISP tasks. While this approach has shown performance gains in outdoor scenarios, it faces challenges in indoor conditions due to factors such as artificial lighting flicker. These findings, along with their implications, have been discussed in greater detail in the revised manuscript and supplementary materials.

Q.7. Experimental Visual Results and Layout

Thank you for highlighting this. We have significantly expanded the visual results in the supplementary materials. This includes: Adding more examples, particularly for low-light and fast-motion scenarios, as shown in Figure 10. Adjusting the layout in the main paper to ensure that visual results are presented more clearly and intuitively.

Q.8. Sample-Specific Analysis

We appreciate this suggestion. In response, we have conducted additional analyses in the supplementary materials to evaluate the performance of different dataset subsets. This provides deeper insights into the dataset's characteristics and potential applications.

Q.9. Task-Specific Effects in the Dataset

Thank you for raising this question. In the supplementary materials (Section E), we have added analyses discussing task-specific effects, such as the impact of lighting flicker in indoor scenes. These findings provide valuable insights into the dataset's strengths and limitations in supporting diverse ISP tasks.

Q.10. Comparison with Existing Methods

We acknowledge that the proposed method is a baseline approach designed to demonstrate the utility of the dataset rather than achieve state-of-the-art performance. The method involves a simple integration of event data into UNet and serves as a foundation for future research. We have included additional discussions in the supplementary materials analyzing its limitations, such as challenges in modeling global context, and its potential for further improvement.

Once again, we thank you for your constructive feedback. Your comments have been instrumental in refining the quality and clarity of our work. We hope our responses address your concerns comprehensively and look forward to your further insights.

Sincerely,

Thank you for your valuable suggestions. In the revision paper, we have incorporated additional visual examples to intuitively showcase the specific content of the dataset. We have also improved the formatting to make the paper's structure more logical and accessible. Furthermore, we have expanded discussions on dataset scale and background knowledge. We hope you have had the chance to review these updates and look forward to engaging in further discussions with you.

Dear Reviewer q8r6,

Thank you again for your valuable comments. We have carefully addressed your feedback and updates into the revision paper. We sincerely hope to know if there are any unresolved concerns and look forward to engaging in further discussions with you.

Specifically, we have made the following revisions to address your points:

1. Sec. B [Q.1, Q.2]:

• Added additional visual examples (Fig. 11 and Fig. 13) showcasing various dataset scenarios.
• Included qualitative analyses (Fig. 12) summarizing key elements, lighting conditions, and their distributions.

2. Sec. A [Q.3]:

• Expanded discussion on the imaging principles of events and frames, emphasizing the advantages of event data for ISP tasks.
• Clarified the evaluation of the controllable ISP framework, which relies on ColorChecker-based assessments (Fig. 5). While intermediate processes like demosaicing and white balancing lack ground truth for quantitative analysis, we provided an overall pipeline performance evaluation.

3. Fig. 6 and Fig. 7 [Q.4]:

• Revised these figures as per your suggestion to reduce redundancy and improve clarity.

4. Tab. 8 and Sec. F [Q.5]:

• Conducted a comprehensive quantitative analysis of overall dataset performance.

5. Sec. 5.1, 5.3, D, and F [Q.6]:

• Discussed the benefits and challenges of event-data fusion, including its advantages and limitations under artificial lighting.

6. Fig. 6, Fig. 7, and Fig. 9 [Q.7]:

• Improved layout and added visual examples to illustrate the unique strengths of hybrid vision sensors.

7. Fig. 6, Fig. 7, Fig. 19 and Fig. 20 [Q.8]:

• Showcased the performance of the same method under varying scenes, providing insights into dataset diversity.

8. Sec. 5 and E [Q.9]:

• Analyzed task-specific performance across indoor and outdoor scenes, with a focus on lighting conditions and event-data gains.

9. Sec. 5.1, 5.2, and F [Q.10]:

• Provided a simple baseline fusion method, discussing its potential gains and associated challenges for future event-based ISP research.

We sincerely appreciate your time and effort in evaluating our work and look forward to engaging in further discussions to clarify any outstanding concerns.

Best regards, ICLR-4440 Authors