Evaluating the Point Cloud of Individual Trees Generated from Images Based on Neural Radiance Fields (NeRF) Method
Round 1
Reviewer 1 Report (Previous Reviewer 3)
Comments and Suggestions for AuthorsThe paper provides a comparison between an image-based deep learning method (Neural Radiance Fields, NeRF) and 3D Reconstruction based on photogrammetry (using Multi-View Stereo, MVS in COLMAP).
Both approaches are based on accurate pose estimation with Structure from Motion (SfM) [7]. Terrestrial laser scanning (TLS) is used as ground truth.
The study shows that NeRF can provide more accurate reconstructions of trees, in particular in the canopy area. Also, the reconstruction of branch structures based on the AdTree [25] branch model yields very good results.
The paper presents original research with compelling examples. It appears to be scientific sound, just reproducing the algorithm is difficult as it relies on many tools. Overall, an interesting paper.
Comments on the Quality of English LanguageWell written, just minor issues. Example:
line 415: "Dense point cloud generated from these two methods was compared to the reference point cloud obtained ...": Here, I would use plural.
Author Response
Thank you for your comments and suggestion. We changed the sentence you mentioned accordingly to: Dense point clouds generated from these two methods were compared to the reference point cloud obtained from multi-station terrestrial laser scanning for reconstruction completeness and quality analysis.
Reviewer 2 Report (Previous Reviewer 2)
Comments and Suggestions for AuthorsThank you for addressing my comments.
Author Response
Thanks again for helping making this manuscript better.
Reviewer 3 Report (Previous Reviewer 1)
Comments and Suggestions for AuthorsThe authors have effectively responded to the reviewers' suggestions, and this paper has been resubmitted, most notably by emphasizing the innovative parts of their work more explicitly.
Author Response
Thank you for your comments.
Reviewer 4 Report (New Reviewer)
Comments and Suggestions for AuthorsThe paper presents a new comparative approach to the methods of detection and calculation of images and terrain data from photogrammetry and remote sensing, with applicability in the study of tree canopies.
The literature review part is insufficient and must be improved with relevant bibliographic references, which reveal the necessity of this approach presented in the case study.
Why did the authors choose some two trees for the case study? It is necessary to explain the appropriateness of choosing this number of trees, at the expense of a larger and more representative number.
It is necessary to present in more detail the relevant technical characteristics of the equipment chosen for field data collection.
Author Response
Comment 1: The paper presents a new comparative approach to the methods of detection and calculation of images and terrain data from photogrammetry and remote sensing, with applicability in the study of tree canopies.
Response 1: Actually our manuscript compared the 3D point cloud of trees derived from overlapping images using traditional photogrammetric method and the newly-developed neural network method. The foci were on the quantitative and qualitative analyses of tree point clouds. We presented different methods for image processing, but didn’t discuss methods of detection and calculation of terrain data.
Comment 2: The literature review part is insufficient and must be improved with relevant bibliographic references, which reveal the necessity of this approach presented in the case study.
Response 2: The research field is developing rapidly and an exhaustive review is impossible and unnecessary for the scope of this study. We did try to include all relevant references in our manuscript. Also we made some modifications in the Introduction: Since the original NeRF was proposed, the field has grown explosively with hundreds of papers extending or building on it each year, and this method has found new applications in various areas, including autonomous driving [12], medicine [13], digital human body [14], 3D cities [15], and cultural heritage reconstruction [16,17], just to name a few.
Comment 3: Why did the authors choose some two trees for the case study? It is necessary to explain the appropriateness of choosing this number of trees, at the expense of a larger and more representative number.
Response 3: We chose two trees with distinct canopy features for comparison, as we learned from previous studies that current photogrammetry methods perform differently for these trees. As we mentioned in the Introduction, “Current photogrammetry method can deal with trees with prominent stem and branches features that are not covered or concealed by the foliage, but for trees with dense leaves in their canopies, the reconstruction result is often not satisfactory”.
We further made some change in Section 2.1 Study Area: We chose two trees with unique features located within the Qishan Campus of Fuzhou University as representative examples for this study.
Comment 4: It is necessary to present in more detail the relevant technical characteristics of the equipment chosen for field data collection.
Response 4: Thanks for the comment. We added more information about the technical characteristics of the equipments used for data collection in Section 2.3.1.: “These different types of consumer-grade cameras are accessible and widely used; the image resolution, numbers of acquired images are shown in Table 1. Considering the need for quality ground truth data, a RIEGL VZ-400 terrestrial laser scanner (main specifications: ranging accuracy 3mm, precision 5 mm, laser beam divergence 0.35 mrad) was used…”
Once again, thank you for your comments and suggestion.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe literature review requires expansion, and I've noticed a technical issue resulting in missing citation numbers in the paper.
Upon my analysis, it appears that the concept under discussion lacks novelty, which diminishes the need for an extensive report on my part.
In the field, numerous studies have explored the use of NeRT in point clouds. For instance, the article by Remondino, F.; Karami, A.; Yan, Z.; Mazzacca, G.; Rigon, S.; Qin, R., titled 'A Critical Analysis of NeRF-Based 3D Reconstruction' (Remote Sens. 2023, 15, 3585. https://doi.org/10.3390/rs15143585), offers substantial content in this area. Regrettably, this significant paper has also been overlooked in the citations.
This particular paper provides a detailed comparison between NeRF-based techniques and standard photogrammetry methods, such as COLMAP. It includes an evaluation of various state-of-the-art methods, with Instant-NGP highlighted for its exceptional results, offering a thorough assessment of NeRF's performance relative to COLMAP.
Additionally, there is another relevant paper focusing on cultural heritage, accessible here: https://ui.adsabs.harvard.edu/abs/2021ISPAr43B2..565C/abstract.
However, theoretically, the paper in question does not introduce significant novelty in its methods. While I recognize the authors' efforts, the content, due to its lack of theoretical innovation, does not seem to meet the standards necessary for publication in a high-impact journal.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors,
The study is very interesting and worth publishing. I consider that moderate improvements are needed:
1. The output quality of Colmap tool is unknown. A thorough comparison with the other photogrammetric software is recommended to confirm the conclusions.
2. I have doubts about the UAV data due to the following statement: "Concerned that open-sourced COLMAP might not be powerful enough, we also tested these two image datasets in the popular commercial photogrammetry software Metashape (Version 2.0.3) and Pix4dMapper (Version 4.7.5), getting the same failed reconstruction results." An issue with the UAV block configuration might exist. A sound comparison is only possible with reliable datasets.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper compares photogrammetric methods with Neural Radiance Fields (NeRF) used for reconstructing trees from camera images. Results based on different error metrics, branch strucures and computation times are presented.
For publication, the following issues sould be fixed:
1. There are lots of dead links to references ("Error! Reference source not found")
2. The techiques compared like multi-view stereo (MVS) and NeRF should be explained in detail. How are these implemented and which parameters have been used? Also, the branch reconstruction may be explained.
3. Metrics like mean cloud distance and cloud-to-cloud distance should be defined. Also, some mathematical notation explaining the approaches is higly appreciated.
4. How do the results compare to the literature, for example [11]?
Further issues:
Can the data sets be publised for use as a benchmark?
Do there exist further metrics? For example, the point clouds can be projected back into the image perspectives and compared in 2D.
Comments on the Quality of English Language
All in all, good English writing. Sometimes, plural may be used (or an article like "a", "the"). Example:
17 "... are compared with point cloud" -> "with point clouds" or "with a point cloud"
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have effectively highlighted the novelty of their approach. However, I concur with the one other reviewer's observation regarding Colmap. The output quality of the Colmap tool remains uncertain. To substantiate the conclusions drawn, a comprehensive comparison with other photogrammetric software is recommended
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors,
Thank you for your explanations in the reviewer response document. However, they do not appear in the manuscript, which reduces the scientific soundness of the manuscript.
Kind regards
Author Response
Comment 1: Dear Authors, Thank you for your explanations in the reviewer response document. However, they do not appear in the manuscript, which reduces the scientific soundness of the manuscript.
Response 1: Thank you for your comment. We are sorry that you didn’t get the chance to see where the modifications were made in the revised manuscript; we did uploaded a version containing all the revised contents, unfortunately it didn’t show up at the right place. Now we presented an updated revision, which shows the changes highlighted, followed by the final clean version of the manuscript.
