From Photogrammetry to Virtual Reality: A Framework for Assessing Visual Fidelity in Structural Inspections

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper presented discusses the method for use of VR models for structural inspection of buildings. The paper is very well written and organized. I have only minor comments. I really like the content of it, and I think it can have significant applications.
- At some points the images might be better placed for to appear on the page where they are first mentioned. E.g., figure 3. I know it is hard to organize a manuscript this big for all of them to be on appropriate pages.

- The method for getting related work is interesting. There is not much related work on the topic found though. Maybe some "manual" related work searching in the references of found papers or the papers which referenced found papers could yield additional papers of interest.

- Maybe a short paragraph on discussing how the whole process could be automated would be useful. 

Overall, I think this is a great paper.

 

Author Response

We thank the reviewer for their feedback on improving the quality of our manuscript. We provide detailed, point-by-point responses to each of the reviewer’s comments below. The revised content has been shown in red text in the updated version of the manuscript.

Comment 1: At some points the images might be better placed for to appear on the page where they are first mentioned. E.g., figure 3. I know it is hard to organize a manuscript this big for all of them to be on appropriate pages.

Response 1: We have revised Figure 3 and examined all other figures in the paper, ensuring all figures will appear on the page where they are first mentioned. Please see the revised paper for the updated figures.

Comment 2: The method for getting related work is interesting. There is not much related work on the topic found though. Maybe some "manual" related work searching in the references of found papers or the papers which referenced found papers could yield additional papers of interest.

Response 2: We have adopted the suggestions from the reviewer and explored additional literature papers. This included both backward citation tracking (reviewing the references cited in the studies listed in Table 1) and forward citation tracking (examining papers that cited those studies listed in Table 1). As a result, we have reviewed six additional papers. The work reported in these six additional papers does not change the research gap we previously identified in this study.

Please see the last two paragraphs in Section 2 in the revised manuscript regarding the methodology we used for searching additional papers, and the review of these papers.

Comment 3: Maybe a short paragraph on discussing how the whole process could be automated would be useful. 

Response 3: We have added a discussion of potential automation of the proposed methodologies, which can be found in the first paragraph under Section 8.3 from the updated manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

I think VR is emerging area as per visual inspections are concerned and is a  good tool.

-The authors have described the workflow in a good manner covering all the details.

I recommend authors to consider some minor points to improve the manuscript.

-Limitations can also be stated as VR is not easy for all structures. The benefits at current state of the art are not that much as people who work in the field wont go though such workflow.

-Some parts can be reduced as it is too much detail. I was also expecting some comparision of proposed work with Related Work section. As some people have done VR for bridges and what authors have done new can be stated.

-Also |I felt the gap section is overstated and not all points are achieved .I do not agree that merely visual fidelity will solve all problems as authors are saying. please check.

-first is that vr is being used to asssit in inspections for buildings. I would consider how authors can mention about if AI can be used in combination of VR? can inspectors manually annotate damages and put in the VR assisted tour that authros are talking about. Authors can go through the paper on Artificial intelligence-assisted visual inspection for cultural heritage to address some of the points raised. 

-Some researchers on integration of VR and AI can be stated which authors can do as I am not recomminding any particular ones.

-Can NDT data be integrated with this VR system? because it is also important in inspections.

-I work in the field and have done plenty of inspections .. I also doubt about the economy aspects and its field applications.. PLease keep in mind that point. 

-Some figures can be combined . I do not see reason why Figure 15. Image rendering setup in Blender and some others are separate ...

I recommend minor revisions. The work is novel and is well preesented and will be useful once the VR tech is scalable in the future.

Author Response

We thank the reviewer for their feedback on improving the quality of our manuscript. We provide detailed, point-by-point responses to each of the reviewer’s comments below. The revised content has been shown in red text in the updated version of the manuscript.

Comment 1: Limitations can also be stated as VR is not easy for all structures. The benefits at current state of the art are not that much as people who work in the field wont go though such workflow.

Response 1: We have added the discussion where our method would fail for reconstructing the 3D models for structures with highly obstructed surfaces or minimal visual detail.

Please see Section 8.2 in the revised manuscript for these changes.

Comment 2: Some parts can be reduced as it is too much detail. I was also expecting some comparision of proposed work with Related Work section. As some people have done VR for bridges and what authors have done new can be stated.

Response 2: Regarding the comparison of our work with existing work, it is challenging to perform comparison of our method against existing VR work in the literature due to: 1) there is no existing work in the literature for assessing visual fidelity of the VR models; and 2) most existing VR work do not release their VR model to the public, leading difficulties for implementing our VR assessment methodology to these VR models in the literature.

Please see the first paragraph in Section 5 in the revised manuscript to reflect these aspects.

Regarding the novelty of our method, most of the existing work reviewed in Section 2 focuses on VR user interface development, sensing data integration, and user collaboration. While there is little research that focuses on visual fidelity assessment of VR models. The visual accuracy of model textures was assumed rather than evaluated. To the best of our knowledge, this is the first study to rigorously quantify the visual fidelity of VR models in the context of civil infrastructure inspection.

Please see the last paragraph in Section 3 and the last sentence of the 2^nd paragraph in Section 8.1 in the revised manuscript to reflect these aspects.

Comment 3: Also |I felt the gap section is overstated and not all points are achieved .I do not agree that merely visual fidelity will solve all problems as authors are saying. please check.

Response 3: We have revisited Section 3, and softened our claims. In addition, we also added a new “Section 4 Scope of This Study”, to clarify our scope, and identify tasks that are within VR in structural inspection but outside the scope of this study.

Please see the 4^th paragraph in Section 3 and the newly added Section 4 in the updated manuscript to reflect these changes.

Comment 4: first is that vr is being used to asssit in inspections for buildings. I would consider how authors can mention about if AI can be used in combination of VR? can inspectors manually annotate damages and put in the VR assisted tour that authros are talking about. Authors can go through the paper on Artificial intelligence-assisted visual inspection for cultural heritage to address some of the points raised. 

Response 4: We have added a discussion on how AI could be integrated with VR in the context of this study in the last section of the paper.

Please see the first paragraph in Section 8.3 in the updated manuscript to reflect these changes.

Comment 5: Some researchers on integration of VR and AI can be stated which authors can do as I am not recomminding any particular ones.

Response 5: We have cited references in these areas mentioned by the reviewer.

Please see the first paragraph in Section 8.3 in the updated manuscript to reflect these citations.

Comment 6: Can NDT data be integrated with this VR system? because it is also important in inspections.

Response 6: Our current study focuses on assessing the visual fidelity of the VR model. We did not investigate NDT data integration in this study.

The integration of NDT data into the VR user interface is an interesting strategy and shall be worth future investigations. To integrate NDT or other sensing data into our VR development methodology, one possible solution is to upload this data into the VIAR360 platform, allowing users to read and interact with these datasets.

Please see the newly added Section 4 on the discussion of NDT data and the disclosure of the scope of this study which does not include NDT data integration.

Comment 7: I work in the field and have done plenty of inspections .. I also doubt about the economy aspects and its field applications.. PLease keep in mind that point. 

Response 7: We have added a discussion to acknowledge the limitations of the practical usage of the proposed method.

Please see Section 8.2 in the updated manuscript to reflect these changes.

Comment 8: Some figures can be combined . I do not see reason why Figure 15. Image rendering setup in Blender and some others are separate ...

Response 8: In the first-round review, we used our paper template. For this second around review, we have changed it into the Sensors’ paper template, where texts are not wrapped around the figure anymore. We believe that now the figures and texts are balanced on most of the pages. Fig. 15 (now Fig. 17) is challenging to combine with its adjacent figures (e.g., Fig. 16 or Fig. 18), as these figures focus on different topics. We welcome any additional comments from the reviewer, and will revise our figure layouts accordingly.

Reviewer 3 Report

Comments and Suggestions for Authors

The problem of inspection of engineering structures, especially those that are difficult to access for direct inspection, is relevant, therefore, the presented study is in demand. Using UAV imagery for monitoring purposes is quite reasonable. The transition to virtual reality for engineering control is a modern solution that requires technology development and analysis of the results obtained. The article clearly and consistently describes an algorithm for using the fusion of photogrammetry and virtual reality technologies to evaluate building structures. The reviewer supports this approach, but it seems to us that the further integration of this technology will depend solely on the ability of users to accept virtual reality as a tool in everyday life.

The reviewer has several comments that generally do not reduce the importance of the article.

1.Despite the fact that the aerial surveying conditions have already been published earlier, it would be possible to duplicate the key points:

What is the average pixel size of the original images?;

What size of defects were expected to be identified?

How was the coordinate system of the original image set?

2. It is debatable whether the original images can be considered ideal for detecting geometric distortions. They contain lens distortion errors and perspective distortions. In this case, from the point of view of geometry, the generated images are more reliable. It is possible that for further experiments it is better to compare not the original images, but the corrected ones for distortion.
3. What function exactly did you use to calculate geometric transformations in the mutual orientation of the original and generated images? The fact is that the parameters of the function largely depend on the uniformity of the distribution of connecting points over the image field.
4. It is quite obvious that the deterioration of visual quality is associated with smoothing and interpolation. In this case, it is possible to use algorithms to improve the sharpness.

Author Response

We thank the reviewer for their feedback on improving the quality of our manuscript. We provide detailed, point-by-point responses to each of the reviewer’s comments below. The revised content has been shown in red text in the updated version of the manuscript.

Comment 1: Despite the fact that the aerial surveying conditions have already been published earlier, it would be possible to duplicate the key points:

What is the average pixel size of the original images?;

What size of defects were expected to be identified?

How was the coordinate system of the original image set?

Response 1: Regarding the pixel size, during the field visit and bridge reconstruction, we did not intentionally measure the Ground Sampling Distance (GSD). As shown in Fig.3, the distance between the UAV camera and the bridge façade varies during field UAV image collection. We did not preprogram the UAV for collecting images of the bridge façade due to the site restrictions (limited flight space). The UAV, instead, was manually controlled for collecting bridge façade images. After receiving the reviewer’s report, we tried a few rough estimate methods for calculating GSD, and they yield a range between 0.5 mm/pixel to 1 mm/pixel. Because it is challenging to calculate accurate GSD, we did not include GSD values in the manuscript.

Regarding the coordinate system of the image set, we used a local coordinate system without georeferencing. We have clarified this in the first paragraph under Section 6.1.1 in the updated version of the manuscript.

Regarding the size of the defect, we would like to clarify two points: 1) the primary objective of this study is to develop and evaluate a visual fidelity assessment framework that quantifies how closely a VR model replicates the surface appearance of a real-world structure. The focus is not on the detection or identification of specific defects, but rather on the underlying visual accuracy of the reconstructed model. 2) The size of the defect is also context-dependent. The detectability of structural defects is inherently linked to the physical scale of the structure and the spatial resolution of the VR model. For instance, a large-scale VR model of an entire building may not render small cracks clearly due to limited pixel representation, whereas a VR model of a localized structural element (e.g., a small façade section) could provide sufficient detail for visualizing finer defects.

Please see the newly added “Section 4 Scope of This Study” in the updated version of the manuscript for the discussion of the scope of this study. In this section, we clarified that a few tasks within VR-based inspection, such as defect detection, are out of the scope of this study.

Comment 2: It is debatable whether the original images can be considered ideal for detecting geometric distortions. They contain lens distortion errors and perspective distortions. In this case, from the point of view of geometry, the generated images are more reliable. It is possible that for further experiments it is better to compare not the original images, but the corrected ones for distortion.

Response 2: We agree with the reviewer that field-captured UAV images may exhibit both lens distortion and perspective effects, which can potentially affect the assessment. Applying lens distortion correction and camera calibration could improve our results. As future work, we plan to explore this possibility by incorporating distortion-corrected imagery to further refine the evaluation framework.

We have added a new paragraph to disclose this limitation and discuss solutions to mitigate lens distortion. Please see the last paragraph in Section 8.1 in the updated manuscript for these changes.

Comment 3: What function exactly did you use to calculate geometric transformations in the mutual orientation of the original and generated images? The fact is that the parameters of the function largely depend on the uniformity of the distribution of connecting points over the image field.

Response 3: We use projective transformation (estimateGeometricTransform in MATLAB) to recover the rendered image into the coordinate system of the ground truth image. The success of this transformation relies on the large volume of matched features detected from both ground truth and rendered images in the earlier phase, as stated in the manuscript.

Please see the third paragraph under Section 2.2.2 from the updated manuscript for these changes.

Comment 4: It is quite obvious that the deterioration of visual quality is associated with smoothing and interpolation. In this case, it is possible to use algorithms to improve the sharpness.

Response 4: We agree with the reviewer that a contributing factor to visual discrepancies reported in this study is the inherent smoothing and interpolation during photogrammetric reconstruction, which can reduce the sharpness of fine surface features. While we did not apply post-processing to enhance surface sharpness of the VR model in this study, edge-preserving sharpening algorithms could potentially improve local feature clarity in the rendered VR models.

Please see the last paragraph under Section 8.1 in the updated manuscript for this discussion.

Reviewer 4 Report

Comments and Suggestions for Authors

In this manuscript, the authors present a comprehensive two-phase framework combining UAV-based photogrammetry and VR development with an objective fidelity assessment methodology to evaluate the accuracy of VR models in representing real-world structural conditions. The topic is timely and relevant, particularly in civil engineering contexts where immersive remote inspections are gaining traction. The manuscript is well-structured and clearly articulates both the technological workflow and evaluation strategy, with validations on two diverse case studies. However, while the work is methodologically sound and practically oriented, this reviewer finds several areas that limit its theoretical and broader scientific contributions. The following points are recommended for revision:

1. While the authors define visual fidelity as a quantifiable, technology-centred concept (distinct from subjective authenticity), the scientific rationale behind the selected evaluation methods (e.g., deviation maps, histogram matching) lacks theoretical underpinning. For instance, why were pixel-wise intensity deviations prioritized over feature-based similarity metrics that might be more robust to lighting inconsistencies?  
2. The paper successfully validates the framework on two structures; however, it lacks a discussion of scalability across more complex or urban infrastructure. Consideration of practical deployment challenges, such as computational demands, UAV regulatory constraints, or multi-user VR access, would strengthen the manuscript’s practical impact.
3. The paper’s key contribution, the image-based visual fidelity metric, remains difficult to evaluate in terms of its accuracy, sensitivity, and robustness. Questions that should be addressed include: How does this deviation mapping perform under varied lighting, texture, or resolution conditions? Could the authors simulate or incorporate controlled synthetic environments (e.g., using 3D models with known ground truth) to assess the accuracy of their method? Moreover, the manuscript qualitatively discusses reconstruction limitations, but the scientific impact of these issues is not quantified
4. How sensitive is the final deviation map to errors in image registration? A sensitivity or ablation study showing the contribution of each step (e.g., camera alignment, histogram normalization) would provide greater transparency and replicability.
5. The manuscript positions itself as bridging a gap in structural inspection, but from a computer vision or VR perspective, the novelty is limited. From a scientific perspective, the visual fidelity evaluation method would benefit from a more formalized algorithmic treatment, for example, by framing the assessment as an optimization or learning problem rather than relying solely on image registration and pixel-wise deviation, it can also benefit from validating the fidelity maps against human perception
6. While the paper acknowledges the inherent limitations of photogrammetry, it stops short of quantifying the impact of these errors on visual fidelity outcomes. A comparison of VR reconstructions using alternative methods would contextualise the benefits and trade-offs more clearly.
7. The study strongly focuses on fidelity as a machine-centred objective metric but omits user-centred assessments (e.g., usability, decision-making support). Even a brief mention of potential future user studies or evaluation protocols would help bridge this gap and underscore the system's usability in actual inspections.
8. Although the study emphasizes VR model fidelity, it provides little scientific insight into how fidelity translates into effective inspection outcomes: How does deviation magnitude correlate with missed or false positive detections in real inspections? Is there a threshold of visual fidelity below which VR-based inspection becomes unreliable? These links should be at least conceptually discussed to clarify the implications of the proposed fidelity framework.

Author Response

We thank the reviewer for their feedback on improving the quality of our manuscript. We provide detailed, point-by-point responses to each of the reviewer’s comments below. The revised content has been shown in red text in the updated version of the manuscript.

Comment 1: While the authors define visual fidelity as a quantifiable, technology-centred concept (distinct from subjective authenticity), the scientific rationale behind the selected evaluation methods (e.g., deviation maps, histogram matching) lacks theoretical underpinning. For instance, why were pixel-wise intensity deviations prioritized over feature-based similarity metrics that might be more robust to lighting inconsistencies?  

Response 1: We have added a discussion to provide reasoning for selecting the proposed deviation mapping and histogram matching. Please see the second and last paragraphs under Section 6.2.3 from the updated manuscript for this discussion.

Comment 2: The paper successfully validates the framework on two structures; however, it lacks a discussion of scalability across more complex or urban infrastructure. Consideration of practical deployment challenges, such as computational demands, UAV regulatory constraints, or multi-user VR access, would strengthen the manuscript’s practical impact.

Response 2: We have added a discussion in Section 8.2 to disclose the limitations of our method in practical applications.

In addition, we also added a new section, Section 4 Scope of This Study, to clarify the scope of this paper. Some tasks within VR-based inspection, such as multi-user collaboration in the VR user interface, are outside the scope of this study.

Comment 3: The paper’s key contribution, the image-based visual fidelity metric, remains difficult to evaluate in terms of its accuracy, sensitivity, and robustness. Questions that should be addressed include: How does this deviation mapping perform under varied lighting, texture, or resolution conditions? Could the authors simulate or incorporate controlled synthetic environments (e.g., using 3D models with known ground truth) to assess the accuracy of their method? Moreover, the manuscript qualitatively discusses reconstruction limitations, but the scientific impact of these issues is not quantified

Response 3: We have added a discussion to our paper to cover these concerns raised by the reviewer and disclose the limitations of our method. Please see Section 8.2 in the updated version of the manuscript.

Comment 4: How sensitive is the final deviation map to errors in image registration? A sensitivity or ablation study showing the contribution of each step (e.g., camera alignment, histogram normalization) would provide greater transparency and replicability.

Response 4: We have added a discussion to our paper to cover these concerns raised by the reviewer and disclose the limitations of our method. Please see Section 8.2 in the updated version of the manuscript.

Comment 5: The manuscript positions itself as bridging a gap in structural inspection, but from a computer vision or VR perspective, the novelty is limited. From a scientific perspective, the visual fidelity evaluation method would benefit from a more formalized algorithmic treatment, for example, by framing the assessment as an optimization or learning problem rather than relying solely on image registration and pixel-wise deviation, it can also benefit from validating the fidelity maps against human perception

Response 5: We appreciate these insights from the reviewer. These suggestions, such as adding layers of optimization or machine learning on top of our proposed fidelity assessment method, would be beyond the scope of our study. These ideas would need extensive efforts to articulate and verify. This would be the scope of our future work.

In terms of human perception (user study), we have discussed this as our future work. Please see Section 8.3 in the updated manuscript for the discussion.

Comment 6: While the paper acknowledges the inherent limitations of photogrammetry, it stops short of quantifying the impact of these errors on visual fidelity outcomes. A comparison of VR reconstructions using alternative methods would contextualise the benefits and trade-offs more clearly.

Response 6: The scope of this study is to propose and validate a visual fidelity assessment framework within photogrammetry-based VR models. In our view, it is not relevant to compare our framework in assessing VR models that are not built by photogrammetry (say, by other reconstruction methods).  To the best of our knowledge, as also reviewed by the authors in Section 2, other VR reconstruction methods include using LiDAR or 360 media (say 360-degree images). Our proposed assessment framework is not built for assessing VR models created by these methods. Lastly, it would be practically challenging for our team to revisit testbeds to collect additional datasets due to restrictions on securing LiDAR equipment and traveling to these sites.

Comment 7: The study strongly focuses on fidelity as a machine-centred objective metric but omits user-centred assessments (e.g., usability, decision-making support). Even a brief mention of potential future user studies or evaluation protocols would help bridge this gap and underscore the system's usability in actual inspections.

Response 7: In our first round of revision, we have mentioned the future perspective for the user study. Please see the paragraph below (now under Section 8.2):

“Future work should also expand toward a more human-centered and interdisciplinary assessment of VR-based structural inspection. While this study focused on the visual fidelity evaluation of VR models using computational, algorithm-driven frameworks, the psychological and experiential impacts of VR for end users (e.g., inspectors) remain largely unexplored within civil infrastructure inspection in the current literature. In related domains involving inspection-like tasks outside the civil engineering field, VR has been reported to offer several psychological benefits, such as creating immersive 3D environments, reducing psychological stress, and enhancing model visualization. Nevertheless, these findings have not been systematically investigated in the context of structural inspection. Future research should therefore conduct controlled experiments comparing VR-based and traditional inspection workflows to evaluate user performance, decision-making accuracy, and cognitive load.”

Comment 8: Although the study emphasizes VR model fidelity, it provides little scientific insight into how fidelity translates into effective inspection outcomes: How does deviation magnitude correlate with missed or false positive detections in real inspections? Is there a threshold of visual fidelity below which VR-based inspection becomes unreliable? These links should be at least conceptually discussed to clarify the implications of the proposed fidelity framework.

Response 8: Regarding translating Fidelity to inspection outcome, we have conceptually discussed that visual fidelity is the prerequisite for a successful VR inspection in our manuscript. See the paragraph below under Section 3, which addresses how fidelity serves as the foundation for inspection outcomes.

“Visual fidelity plays a foundational role in VR-based structural inspections. An inaccurate VR model may distort structural damage (e.g., cracks, corrosion, or material degradation) and lead to incorrect observations. For instance, a structural inspector using a VR model might fail to detect a hairline crack or misinterpret surface rust if the visual representation is not accurately preserved. In this regard, high visual fidelity is a prerequisite for building VR user interfaces. Despite the sophistication of the user interface design, if the underlying VR model fails to depict its real-world counterpart, the VR user interface built later on will not yield satisfactory inspection outcomes.”

Regarding the inquiry about proposing a threshold for the deviation map, we have added a discussion in the third paragraph under Section 8.1 of the updated manuscript to address this concern.

Reviewer 5 Report

Comments and Suggestions for Authors

 

 

Dear Authors, your work is very interesting. Civil structures are traditionally inspected in-situ, which is costly, risky, and prone to error the use of Virtual Reality (VR) technologies is an alternative inspection method. Your study proposes a two-phase methodology: VR Model Development and VR Model Assessment.

There are many contributions and advantages like a novel framework for VR user interface development, integrating Blender and VIAR360 for immersive scene construction. This offers a user-friendly, web-based alternative to common VR development tools like Unity. It introduces an innovative methodology for objectively evaluating the visual fidelity of VR models using computational algorithms, rather than relying solely on subjective user perceptions of "authenticity." This fidelity assessment is crucial for ensuring the accuracy of VR models in structural inspections, as inaccurate models can lead to overlooked damage or misinterpretations. The computational nature of the visual fidelity assessment allows for scalable evaluation across numerous models and over time, providing a practical quality control mechanism for civil structure asset management. 

In conclusion the study successfully proposes and validates a comprehensive framework for both developing and assessing the visual fidelity of VR models for structural inspection. It addresses a significant research gap by providing an objective method to ensure VR models accurately replicate real-world conditions, thereby enhancing the reliability of VR-based structural inspections

Also if a critical question remains: can VR models accurately reflect real-world structural conditions? 

Finally some specific areas within the VR models exhibited lower visual fidelity. This was attributed to limited UAV (Unmanned Aerial Vehicle) image coverage during the initial data collection phase, which subsequently led to inaccuracies in the 3D model reconstruction.

Author Response

We thank the reviewer for their feedback on improving the quality of our manuscript. We provide detailed, point-by-point responses to each of the reviewer’s comments below. The revised content has been shown in red text in the updated version of the manuscript.

Comment 1: Also if a critical question remains: can VR models accurately reflect real-world structural conditions? 

Response 1: We appreciate the reviewer’s observation regarding the central question of whether VR models can accurately reflect real-world structural conditions. This concern is indeed the driving motivation of our study. As described in the Introduction and reinforced throughout the manuscript, we developed a two-phase framework that systematically constructs VR models from photogrammetric data and evaluates their visual fidelity using quantitative image comparison techniques. We further demonstrated that the rendered VR models preserve a high degree of visual consistency with UAV-captured images under matched viewpoints.

Comment 2: Finally some specific areas within the VR models exhibited lower visual fidelity. This was attributed to limited UAV (Unmanned Aerial Vehicle) image coverage during the initial data collection phase, which subsequently led to inaccuracies in the 3D model reconstruction.

Response 2: We agree with the reviewer’s comment that UAV image coverage is another factor that may cause the discrepancy between the ground truth and Blender-rendered images. We have added this discussion in the last paragraph under Section 8.1 from the updated manuscript.

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

Thank you for addressing my previous concerns in the revised version of the paper. The updates comprehensively address the points I raised, and I find the current version to be prepared and suitable for publication.