The Applicability of Two Generative Adversarial Networks to Generative Plantscape Design: A Comparative Study

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

Although it is always interesting to see how algorithms are applied in landscape architecture design, as every situation is unique, the following comments are provided to improve the text.

I kindly ask the authors to reconsider the title. As it stands, it is confusing and unclear. It is unusual to start with “study,” and the title includes “- -” and “take flower borders as an example.” What exactly is meant by this?

In the Introduction section between lines 30—35, you mention the usage of flowering species in urban environments and as borders for “woodlands”, but this is problematic, as the concept and composition of species in urban and rural environments are quite different. Please verify this in the source you cited.

I cannot agree with the authors' statement that "flower borders not only occupy small areas and incorporate a variety of plant species" (lines 35-37). A flower border can consist of just one plant species! Please explain this and provide a source to support your claim.

In lines 38 to 47, it is unclear whether you are referring to urban or rural areas, as the design for these areas differs. Please clarify this and kindly provide appropriate references.

As mentioned between lines 61-63, the aim is not fully understandable. What is the purpose of this study? Is it to encourage landscape architects to use algorithms in the design process instead of traditional software or manual design methods? Please add a clearer explanation of the aim in the introduction, possibly using the text under the subsection “2.1. Analytical Framework.”

Lines 137-146 need more references. Please avoid placing the conclusion in the introduction section.

In the section, Materials and Methods— “2.1. Analytical Framework”, line 156, please explain why is this subsection titled as such. You are discussing the aim of the research. Some parts of this subsection are confusing, particularly regarding the creation of a uniform model. I assume the authors are aware that each habitat has its own natural conditions and that the arrangement and composition of plants cannot be uniform or applicable in the same way everywhere. How is this addressed here? I kindly ask the authors to reconsider the structure of this subsection, as the explanation provided is not clear enough.

Please explain all figures in the text more thoroughly. It is not enough to simply state "as shown in the Figure ".

The paper lacks a clearly defined scientific structure. Does the paper mention a specific site where the methodology has been applied? If not, why was this decision made? The methodology blends objectives and results, and in the results section, only some information is briefly mentioned, which may be better explained in the methodology section.

Please explain the text between lines 185-187 more carefully. Line 185 states, "The object of this study is small-scale plantscape design." Do you have a specific area in mind? If not, how can be applicable an imaginary green area in the landscape to all green areas of different categories and natural landscape conditions?

Line 202: What do you mean by “300 plants”?

Line 226: Please define “plant elements.”

Lines 217 and 218: Please explain which 56 subcategories you are referring to.

Line 228: What do these numbers mean?

Figure 4 is too small and needs more explanation about the entire dataset.

Line 243: Why are the 5° and 10° rotations necessary and why in both clockwise directions?

Line 253 (Figure 5): Please correct this.

In the subsection “2.3.1. Flower Border GAN Model”, the discriminator is mentioned. Could the authors please clarify the main goal of the discriminator? Is the transformation from “Gx” to the “y” image referring to a change from the current design to a new transformed one, which the computer suggests in order to enhance the ecological and aesthetic values of the locality, or is there something else involved? Please provide a clearer explanation.

Lines 338 and 339: Please explain what "100, 300, and 500 rounds of training respectively" means. What is the goal of this experiment?

Table 4: What kind of samples are you referring to? Could you please clarify or explain the samples? Are they from the same category or not? Why is this case?

What do you mean by “the total loss function” (line 350)?

Lines 366—391: What is the purpose of this section? How does it relate to the goals of the paper? What do you want to achieve with this—are you trying to change the current design, or use algorithms to begin the design process?

The Results section is too short to show how the methodology was applied and where. Do you have a specific example site, and how could this be applied to other sites? I kindly ask the authors to explain this in more detail, as the current explanation does not make the scientific or practical purpose of the research clear.

Please expand your reference list with more research papers and articles, especially those from outside your geographic area (e.g., sources from Europe, America, etc.).

The Discussion section lacks a scientific format. It should be shaped in a way that compares the results of this paper with those of other studies. The Conclusions should clearly state whether the objective was achieved by applying the appropriate methodology, what the limitations were, and what questions remain for further research.

 

Best regards.

Author Response

Dear Reviewer,

Thank you very much for your valuable time and insightful comments on our manuscript. We have carefully considered each of your suggestions and have made the necessary revisions to improve the quality of our manuscript. Below, we provide a detailed point-by-point response to your comments, along with the specific changes made in the revised manuscript. Please see the attachment for the revised article.

Comments 1: I kindly ask the authors to reconsider the title. As it stands, it is confusing and unclear. It is unusual to start with “study,” and the title includes “- -” and “take flower borders as an example.” What exactly is meant by this?

Response 1: Thank you for your valuable suggestion. We agree with your opinion and have made the changes. We have removed "—using flower borders as an example" and changed the title of the manuscript to “A comparative study on the applicability of two generative adversarial networks to generative plantscape design.”

Comments 2: In the Introduction section between lines 30—35, you mention the usage of flowering species in urban environments and as borders for “woodlands”, but this is problematic, as the concept and composition of species in urban and rural environments are quite different. Please verify this in the source you cited.

Response 2: Thank you for raising this question. We apologize for the confusion caused by our original text. After reading your comments, we realized that our explanation was not clear enough. The focus of this study is on flower borders in urban areas, and the collected flower border samples are also from urban areas. We have revised the concept of flower borders in lines 32-34 to better express its meaning. According to reference 1, the definition of a flower border by the Royal Horticultural Society is: "A flower border is a dynamic planting scheme that combines perennials, annuals, bulbs, and shrubs to create a visually stunning and ecologically balanced garden feature."

Comments 3: I cannot agree with the authors' statement that "flower borders not only occupy small areas and incorporate a variety of plant species" (lines 35-37). A flower border can consist of just one plant species! Please explain this and provide a source to support your claim.

Response 3: Thank you very much for your valuable comment. We realize that our previous statement was not clear enough and may have caused some confusion. To clarify, our study focuses on "Mixed Flower Borders," which typically contain multiple plant species. As can be seen from the definition provided earlier (lines 32-34), flower borders consist of a mix of plants, not just one species. We have revised the text to make this point clearer.

Comments 4: In lines 38 to 47, it is unclear whether you are referring to urban or rural areas, as the design for these areas differs. Please clarify this and kindly provide appropriate references.

Response 4: Thank you for your valuable comment. We sincerely apologize for any confusion caused by the original description. After reviewing your feedback, we realized that further clarification is needed. The focus of this study is on flower borders in urban areas, rather than those in rural settings. To address this, we have added additional clarification in lines 36-38 to specify the scope and context of our study. The added text is as follows: "The flower border, with its diverse plantings, creates a natural and visually layered effect, making it a key form of urban plant landscaping."

Comments 5: As mentioned between lines 61-63, the aim is not fully understandable. What is the purpose of this study? Is it to encourage landscape architects to use algorithms in the design process instead of traditional software or manual design methods? Please add a clearer explanation of the aim in the introduction, possibly using the text under the subsection “2.1. Analytical Framework.”

Response 5: Thank you for your valuable comment. We sincerely apologize for not fully explaining the goal of our study earlier. To provide a clearer explanation of the research objective, we have added a more detailed description in lines 154-159. The revised text is as follows: “This study proposes an experimental procedure for GAN-based plantscape plan generative design model, focusing on comparing different GAN algorithms to create ecologically viable and aesthetically cohesive flower border plans. The aim is to explore an effective solution to the challenges in plantscape design, ultimately enhancing design efficiency and providing a new tool that could be integrated into existing workflows.”

Comments 6: Lines 137-146 need more references. Please avoid placing the conclusion in the introduction section.

Response 6: Thank you for your valuable comment. We have revised this section accordingly. This part is a summary of the literature review presented earlier, and the relevant references have already been mentioned in the previous section. In response to your suggestion, we have added additional references to further support the argument and strengthen the discussion.

Comments 7: In the section, Materials and Methods— “2.1. Analytical Framework”, line 156, please explain why is this subsection titled as such. You are discussing the aim of the research. Some parts of this subsection are confusing, particularly regarding the creation of a uniform model. I assume the authors are aware that each habitat has its own natural conditions and that the arrangement and composition of plants cannot be uniform or applicable in the same way everywhere. How is this addressed here? I kindly ask the authors to reconsider the structure of this subsection, as the explanation provided is not clear enough.

Response 7: Thank you very much for your valuable feedback. We have made the following revisions and clarifications in response:

1. We realize that the description of our data collection process may have caused some confusion, and we sincerely apologize for any misunderstanding. The samples we collected come from the East China region, which shares similar climate and soil conditions. While the specific locations of the sites vary, all of them are urban and relatively flat. Furthermore, these samples were sourced from two design firms with similar design styles, utilizing comparable plant species and design logic. We believe that these commonalities in design allow us to establish a model for our research. Due to the lack of publicly available datasets on flower border design, we were limited in the number of samples we could collect. Consequently, we were unable to provide a more detailed classification of the environmental conditions of the sites. As a result, the model we have established focuses on exploring flower border design patterns within the broader East China region. We sincerely appreciate your suggestion and will make sure to take it into consideration for future experiments.
2. To make this section clearer, we have carefully revised the content and added further explanations about the overall experimental framework. We have also reworded the introduction to enhance clarity. The changes are as follows (lines 164-165): "This chapter presents the framework for the experimental procedure of the GAN-based plantscape plan generative design model."
3. We have further clarified the standards for dataset collection in lines 167-174 to ensure the scientific validity and rigor of the study. The updated explanation is as follows: Due to the lack of publicly available datasets on flower border design in existing studies, this study collected a series of plan drawing samples from several completed plant configuration projects. To ensure the scientific validity and consistency of the data, all samples were drawn from urban sites in the East China region, which shares similar climate and soil conditions. The sites were selected based on their high design quality, ensuring both a broad representation of site types and design styles, while maintaining rigorous scientific standards.

Comments 8: Please explain all figures in the text more thoroughly. It is not enough to simply state "as shown in the Figure ".

Response 8: Thank you for your valuable suggestion. We have provided more detailed explanations for Figure 2 and Figure 3. The specific chart explanations have been added in the relevant sections. Based on your suggestion, we have enlarged the text in Figure 4 and ensured that the information in the chart is clearer and easier to understand. Additionally, we have made sure that all other images are thoroughly explained in the text. If you have any further suggestions or require additional changes, we would be more than happy to continue making improvements.

Comments 9: The paper lacks a clearly defined scientific structure. Does the paper mention a specific site where the methodology has been applied? If not, why was this decision made? The methodology blends objectives and results, and in the results section, only some information is briefly mentioned, which may be better explained in the methodology section.

Response 9: Thank you for your valuable suggestion. Regarding your question, I would like to clarify the explanation of our research subject and methodology. Although our method was not applied to a specific site, we selected representative site types and tested and evaluated the generated results. We also used modeling to make the test results more intuitive and invited experts to score these results. These steps help validate the effectiveness of our method and provide actionable design solutions.

Regarding the Results section, we recognize that the current structure may have caused some confusion. In the Methodology section, we described the entire experimental process, but some of the results were not discussed clearly or systematically. Based on your suggestion, we have revised the Methodology section to clearly separate the methodology from the results and rewrited the relevant content. We have moved the sections related to the results from the Materials and Methods to the Results, specifically under 3.1. Two Rounds of Test Results and 3.3. Optimization of the Results to present the experimental results and discussion more clearly. We have made substantial revisions to the paper to ensure the Methodology and Results sections are clearer, with additional explanations and details where necessary.

Comments 10: Please explain the text between lines 185-187 more carefully. Line 185 states, "The object of this study is small-scale plantscape design." Do you have a specific area in mind? If not, how can be applicable an imaginary green area in the landscape to all green areas of different categories and natural landscape conditions?

Response 10: Thank you very much for your valuable suggestions. In response to your feedback, we have made the following clarifications and revisions. The term "small-scale" in the original text refers to flower border designs with an area smaller than 300㎡. Although the specific locations of the sample sites vary, they share several important characteristics. First, all sites are in the East China region, ensuring similarities in climate conditions, soil types, and plant species. Second, all sites are urban and feature relatively flat terrain. Finally, the samples come from two design firms with similar styles, which ensures consistency in the plants used and the design logic.

Given these commonalities, we believe the model we established is suitable for similar sites in the East China region. The effectiveness of this approach is based on specific regional and site conditions. It forms the foundation of our study. In future experiments, we plan to include a wider range of site types and natural landscape conditions to further assess the broader applicability of this method.

To improve clarity, I have added the following explanation in lines 197-200: “To ensure consistency and relevance, the sites selected for the plans are all located in the East China region, sharing similar climate conditions, soil types, and plant species. Although the specific locations differ, all the sites are urban and feature relatively flat terrain.”

Comments 11: Line 202: What do you mean by “300 plants”?

Response 11: Thank you very much for your thoughtful question. We sincerely apologize for any confusion caused by our original wording. Upon reviewing your feedback, we realized that our explanation was not sufficiently clear. By "300 plants," we meant that after analyzing the plants in the collected sample plans, we identified a total of 300 distinct plant species used across the plans. We then categorized and merged these 300 plant species based on the criteria mentioned below. We regret that the original wording may have caused confusion, and we have made the following revisions in lines 214-218 to clarify: “After analyzing the plants in the collected plans, we identified 300 distinct and commonly used plant species. To enhance the efficiency of learning typical layout relationships in flower border plant design, we categorized and merged these 300 plant species. This process follows the established plant classification conventions of flower border design and aims to extract key information regarding plant combination patterns.”

Comments 12: Line 226: Please define “plant elements.”

Response 12: Thank you very much for your valuable suggestion. We apologize for the oversight in our original text. The definition of "plant elements" is as follows: “In this study, "plant elements" refer to individual plant patches or groupings within a flower border plan. They are considered distinct components for generative design, based on their type, size, shape, and spatial arrangement. ” We have added this definition in lines 224-227.

Comments 13: Lines 217 and 218: Please explain which 56 subcategories you are referring to.

Response13: Thank you for your valuable suggestion. We apologize for uploading this material in the non-published material file in our previous submission. We have uploaded a table of plant elements to the supplementary materials. It includes the plant elements and their RGB values.

Comments 14: Line 228: What do these numbers mean?

Response 14: Thank you for your question. We apologize for not explaining this clearly in the original text. To make the color differences in the generated images as large as possible, the color differences in the dataset labels should also be as large as possible. This study uses the RGB color model, so we divide the values of the R, G, and B channels into as few parts as possible. Each channel ranges from 0 to 255. When we divide 255 into n parts, we can express (n+1)³ colors. Since there are 56 plant elements in this study, the minimum value of n is 3 to ensure the color labels cover all plant elements. When 255 is divided into 4 parts, we get values of 0, 85, 170, and 255.

Comments 15: Figure 4 is too small and needs more explanation about the entire dataset.

Response 15: Thank you for your valuable suggestion. We have increased the font size in Figure 4. Regarding the explanation of the dataset, we have added information about the dataset’s source and sample size in the figure. Additionally, we have provided a more detailed explanation in the text to help readers better understand the composition and background of the dataset.

Comments 16: Line 243: Why are the 5° and 10° rotations necessary and why in both clockwise directions?

Response 16: We apologize for not explaining this clearly in the original text. In this study, we used data augmentation by rotating the annotated images. The images were rotated 5° and 10° in both clockwise and counterclockwise directions. In computer vision, data augmentation creates diverse image samples through methods like geometric transformations (rotation, flipping, cropping), color transformations (brightness, contrast adjustments), and adding noise. Since flower borders have scale information and require completeness, we chose to rotate the original images. Rotating by 5° and 10° in both directions is a common and effective method, as verified in the relevant literature.

Comments 17: Line 253 (Figure 5): Please correct this.

Response 17: Thank you for your valuable comment. We realized that we missed the numbering, and we have made the necessary revision. The section on CycleGAN has been updated to "2. CycleGAN."

Comments 18: In the subsection “2.3.1. Flower Border GAN Model”, the discriminator is mentioned. Could the authors please clarify the main goal of the discriminator? Is the transformation from “Gx” to the “y” image referring to a change from the current design to a new transformed one, which the computer suggests in order to enhance the ecological and aesthetic values of the locality, or is there something else involved? Please provide a clearer explanation.

Response 18: Thank you for your question. We have added an explanation for the purpose of the discriminator in lines 276-277 “The main goal of the discriminator is to ensure that the generated images are not only realistic but also align with the input conditions.”

Regarding the transformation from G(x) to y, this refers to the process in which the generator creates a flower border design (y) based on the given input (x, the site boundary). This process does not propose new designs to improve ecological or aesthetic value directly. Instead, it generates a design based on the specific conditions of the input site.

The generator learns design patterns and relationships from the dataset. It produces a design that aligns with the input conditions as much as possible. Once training is complete, the generator can create flower border designs based on new input conditions. These designs can be applied to various design scenarios.

Comments 19: Lines 338 and 339: Please explain what "100, 300, and 500 rounds of training respectively" means. What is the goal of this experiment?

Response 19：Thank you for pointing this out. To respond to your question, we have added two explanations in lines 361-363 and lines 342-345 respectively, to clarify the training epochs and the purpose of the experiment.

In this study, " rounds of training" refers to the number of times the model processes the dataset during training , with each round allowing the model to learn and adjust its parameters based on the data. Therefore, 100, 300, and 500 epochs represent the model’s performance at different stages of training.

The goal of this experiment is to determine the optimal number of training iterations at which the model's output achieves the best balance of clarity, realism, and consistency in the plant arrangement.

Comments 20: Table 4: What kind of samples are you referring to? Could you please clarify or explain the samples? Are they from the same category or not? Why is this case?

Response 20: Thank you very much for your valuable question. We sincerely apologize for any confusion caused by our previous explanation. Regarding the samples in Table 4, they primarily represent a type of curved site. However, there are some differences among these sites, which include crescent-shaped, near-circular, and elongated curved forms. These variations are intended to demonstrate the model’s performance in generating designs for different spatial configurations, helping to assess its adaptability and effectiveness in various layouts. We deeply regret not providing this clarification earlier in the paper. In response, we have added the following explanation in lines 371-374: The samples in Table 4 represent different curved site shapes, including crescent, near-circular, and elongated forms. These variations are used to evaluate the model’s performance and adaptability in generating designs for different spatial layouts. We hope this additional explanation helps clarify any uncertainties.

Comments 21: What do you mean by “the total loss function” (line 350)?

Response 21: In the Pix2Pix model, the total loss is a comprehensive metric used to assess the quality of the generated images during the training process. The convergence of the loss function indicates that the training is relatively complete. It consists of two main parts: the generator loss and the discriminator loss. The total loss is typically expressed as: Total Loss = Generator Loss + Discriminator Loss. In the paper, the total loss function refers to the specific changes in the total loss during our experimental training process.

Comments 22: Lines 366—391: What is the purpose of this section? How does it relate to the goals of the paper? What do you want to achieve with this—are you trying to change the current design, or use algorithms to begin the design process?

Response 22: We apologize for not explaining this issue clearly in our previous manuscript. Machine-generated results can be a starting point for design. However, these results often have flaws and cannot be directly used in practical design. Manual optimization is needed to improve the image clarity and ensure they meet the professional standards of flower border design and construction. As mentioned earlier, the goal of using machine-generated images is to support the design process, not to fully replace manual design. This aims to improve design efficiency and quality. The main purpose of this section is to confront the imperfections in the generated results and make them more suitable for actual design work.  We also added clarification in lines 413-416: "The purpose of this section is to confront the imperfections in the automatically generated results, which are not directly suitable for use in design. Therefore, manual optimization is needed to make the images clearer and more usable in practical design applications."

Comments 23: The Results section is too short to show how the methodology was applied and where. Do you have a specific example site, and how could this be applied to other sites? I kindly ask the authors to explain this in more detail, as the current explanation does not make the scientific or practical purpose of the research clear.

Response 23: Thank you for your valuable suggestion. We sincerely apologize for the confusion. We have adjusted the manuscript, especially in the Materials and Methods section. The content related to the results and methods has been moved to the Results section to improve the clarity of the structure.

Regarding the application of the model to specific sites, we did not use a single specific site for the study. Instead, we tested it using a dataset we constructed. The test set includes rectangular and curved sites. The ratio of the test set to the training set is 1:4. This setup allowed us to validate the model’s adaptability and generation performance on these two different site types. Based on these results, we concluded that the model can adapt to various types of site layouts.

We have provided a more detailed explanation of these test results in the Results section. We also demonstrated the model’s performance on different site types. We hope these changes help clarify the scientific and practical significance of our research.

Comments 24: Please expand your reference list with more research papers and articles, especially those from outside your geographic area (e.g., sources from Europe, America, etc.).

Response 24: Thank you for your valuable suggestion. Some of the references in the previous version are from Chinese authors, but they were published in English journals. As a result, there are fewer references from other regions. To address this issue, we have added more references, especially those from other regions.

Comments 25: The Discussion section lacks a scientific format. It should be shaped in a way that compares the results of this paper with those of other studies. The Conclusions should clearly state whether the objective was achieved by applying the appropriate methodology, what the limitations were, and what questions remain for further research.

Response 25: Thank you for your valuable suggestion. 

1. in the Discussion section: GAN algorithms for flower border design, which is a relatively new area of study. As a result, there are fewer available studies for comparison. Additionally, most existing research in plant landscape design tends to concentrate on generating facade renderings, whereas our study specifically focuses on the design of flower border floor plans. This difference in research focus creates a certain gap between our work and existing studies. Given the novelty of this field and the limited relevant literature, we are concerned that making direct comparisons with existing research might not effectively highlight the uniqueness and contributions of our study. Nonetheless, we will make every effort during the revision process to ensure that our discussion clearly emphasizes the innovation and value of our research.
2. in the Conclusions section: We acknowledge that the structure of the Conclusions section may not have been sufficiently clear and did not fully explain how the research objectives were achieved or address the study’s limitations. We have revised this section to ensure that it clearly answers the research questions. In lines686-697, we have made follow changes: “Experiments have shown that both Pix2Pix and CycleGAN are capable of generating garden layout designs. Pix2Pix demonstrates a stronger ability to recognize the site boundary, while CycleGAN generates clearer color blocks that closely resemble real plant patches in flower border designs.

However, the generated results also exhibit certain limitations. In the images gen-erated by Pix2Pix, the color blocks are often fragmented and their distribution tends to resemble that of the training dataset. CycleGAN, on the other hand, shows some shortcomings in accurately identifying site boundaries. For the CycleGAN model, which performed better in the experiment, professionals were also invited to provide an evaluation. The evaluation results indicated that the generated flower border de-sign plans have certain ornamental and ecological value, but there is still room for im-provement in aspects such as vertical variation, texture harmony, low maintenance, and sustainability.”

Best regards.

We believe that the revisions have significantly enhanced the clarity and quality of our manuscript. We have highlighted the changes in the revised manuscript for your convenience.

Thank you once again for your support and guidance. We look forward to your positive response.

Sincerely,

Sun Yuting

Reviewer 2 Report

Comments and Suggestions for Authors

Comments to the Author：

 

*Line 191-200: If conditions permit, the selected dataset can be made public.

 

*The list of selected plant species may be provided in the supplementary materials.

 

*According to the author's description, the author uses the designed plan to train the model (lines 198-200). How can we ensure that the plan represented by color blocks can generate a flower border that is equally beautiful vertically?

 

*For a high-quality flower border designer, is it necessary to use such a complex method to generate the scheme of small-scale flower borders? Can we further elaborate on the application prospect of this technology on a larger spatial scale?

Author Response

Dear Reviewer,

Thank you very much for your valuable time and insightful comments on our manuscript. We have carefully considered each of your suggestions and have made the necessary revisions to improve the quality of our manuscript. Below, we provide a detailed point-by-point response to your comments, along with the specific changes made in the revised manuscript. Please see the attachment for the revised article.

Comments 1：*Line 191-200: If conditions permit, the selected dataset can be made public.

Response 1: Thank you for pointing this out. However, we are sorry that we cannot make the dataset publicly available. The dataset comes from two specific design institutes, and due to privacy concerns, we are unable to share it.

Comments 2： *The list of selected plant species may be provided in the supplementary materials.

Response 2：Thank you for your valuable suggestion. We apologize for uploading this material in the non-published material file in our previous submission. We will re-upload the table of plant elements in the supplementary materials. It will include the specific plant elements and their corresponding RGB values.

Comments 3: *According to the author's description, the author uses the designed plan to train the model (lines 198-200). How can we ensure that the plan represented by color blocks can generate a flower border that is equally beautiful vertically?

Response 3: Thank you for pointing this out. We understand the importance of this issue. Due to the interpretability of the algorithms, we cannot guarantee 100% certainty. However, the dataset we used has been carefully selected and quality-controlled. This ensures that the generated designs meet certain standards in layout, elevation, and other aspects. We have added a clarification in lines 170-174 “To ensure the scientific validity and consistency of the data, all samples were drawn from urban sites in the East China region, which shares similar climate and soil conditions. The sites were selected based on their high design quality, ensuring both a broad representation of site types and design styles, while maintaining rigorous scientific standards.” The design samples in the dataset were rigorously screened to ensure their quality. Additionally, we invited experts to evaluate the generated results. This helps validate and optimize the effectiveness of the designs.

Comments 4: *For a high-quality flower border designer, is it necessary to use such a complex method to generate the scheme of small-scale flower borders? Can we further elaborate on the application prospect of this technology on a larger spatial scale?

Response 4: Thank you for raising this question. The goal of this study is to explore how algorithms can improve the efficiency of the design process, especially for complex, repetitive, or data-intensive design tasks. In such cases, automatically generated designs can serve as a tool for designers. They can help with preliminary design, iteration, and optimization. Additionally, if the model generates ideal layouts after training, designers would use the model's trained results without needing to follow the entire experimental process. Regarding the potential application of this technology at larger spatial scales, we believe it has great potential. With the advancement of technology and continuous optimization of algorithms, this method can be applied to large-scale plant landscape design. For example, in large projects such as urban public spaces or garden landscapes, automated design generation can help designers explore diverse design options more quickly. It can also optimize designs based on ecological, environmental, and aesthetic standards. This method not only increases design efficiency but also provides designers with richer creative inspiration.

Best regards.

We believe that the revisions have significantly enhanced the clarity and quality of our manuscript. We have highlighted the changes in the revised manuscript for your convenience.

Thank you once again for your support and guidance. We look forward to your positive response.

Sincerely,

Sun Yuting

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

Thank you for your manuscript. I have proceeded with corrections to your article, however, 

I doubt about the originality of your paper, since you have published the same article in Research Gate

https://www.researchgate.net/publication/384480815_Generative_Design_of_Plantscape_Based_on_Generative_Adversarial_Network_A_Case_Study_of_the_Generation_of_Flower_Border_Plan

Could you please explain what is the difference with this article from Research Gate?

Please also have a look on my comments to your article, which will help you to improve your manuscri

-In line 39. The authors state that, “Firstly, it is difficult to generalize design rules”. Please elaborate further this argument, since there are many scholars writing about planting design guidelines and general rules exist.

-Line 55-56. There have been a small number of research results on the application of deep learning techniques in the field of landscape architecture, but there is a lack of research on

small-scale plantscape design. Please justify this argument with the appropriate bibliography.

-Avoid using the first singular or plural person in your sentence. Eg. Line 67. We implemented the study according to the following steps

-In the introduction do not describe the methodology that you were following. Lines 67-79. The authors could briefly describe what the readers should expect to find in the following sections.

-The research questions to be at the end of Introduction before the paragraph with the goals.

-From the section materials and methods, please extract any results and add them in the separate section of “Results”.

-The same applies to the “Discussion”. There are a lot of examples that could be added to the Results section. Tables or graphs from the Discussion section should be included in the Results section. Please try to organize your results in the same structure of paragraphs as presented in the Discussion section. For example: 4.1.1. Image Quality, 4.1.2. Design Mode, etc

 

 

Author Response

Dear Reviewer,

Thank you for your attention to our manuscript, A Study on the Applicability of Generative Adversarial Networks to Generative Plantscape Design—Taking Flower Borders as an Example, and for your valuable comments. We highly appreciate the concerns you raised regarding the novelty of our work, and we would like to provide a detailed clarification and supplementary explanation. This study builds upon our previous work. However, it makes a substantial new contribution by offering a comparative analysis of two generative algorithms, an aspect not addressed in the published literature. Below, we outline the three key differences in detail.

1. Core Contribution: Comparative Algorithm Analysis

The core focus of this research lies in evaluating the effectiveness of two generative algorithms (Pix2Pix and CycleGAN) in the context of plantscape design, which constitutes the central innovation of our study. 

Unlike our previous research, which used only CycleGAN, this study compares the results from both CycleGAN and Pix2Pix algorithms. After training, we compared the best results generated by Pix2Pix and CycleGAN, analyzing the strengths and weaknesses of each algorithm in flower border generation. This systematic comparison aims to reveal how different algorithmic approaches influence the applicability of plantscape design.

2. Enhanced Experimental Dataset

To facilitate a more effective comparison of the algorithms, we significantly expanded and optimized our experimental dataset.

In the previously published paper, we created a single dataset for CycleGAN, which contains 60 samples.

In this study, we collected more flower border design samples, increasing the original dataset from 60 to 123 samples. Additionally, we constructed two separate datasets for each algorithm. The first dataset, which includes images rotated 5° in both directions, consists of 369 samples, while the second dataset, with rotations of 10° in both directions, contains 615 samples. As a result, we established four distinct datasets in this study, marking a substantial increase in both quantity and quality compared to our prior research.

3. Advanced Evaluation Framework

In the previously published paper, expert evaluations were based solely on the generated planting plans. 

In the current study, we incorporated Lumion modeling and Importance-Performance Analysis (IPA) as additional evaluation methods. Prior to distributing the questionnaires to the experts, we optimized the generated flower border designs and used Lumion to create 3D models, making the results more intuitive and concrete. In the questionnaires, we asked experts to rate the performance of each criterion (i.e., the "performance degree") as well as the importance of each criterion. More important criteria were prioritized accordingly. After obtaining the expert ratings, we compared the importance of each criterion with the corresponding performance in the generated results, producing IPA quadrant charts to identify areas that should be prioritized for improvement.

Given these innovations, it is possible that the original title did not fully capture the core focus of the study, leading to concerns about its novelty. We believe the following title better reflects the novelty of our work and emphasizes its new research direction: 

A Comparative Study on the Applicability of Two Generative Adversarial Networks to Generative Plantscape Design.

We appreciate the reviewer’s feedback and the identification of these differences, which may not have been sufficiently clear in the earlier version of the manuscript. This study represents an innovative extension of our prior work, particularly in the "Experiment" and "Discussion" sections. We trust that this clarification addresses your concerns regarding the novelty of our research. We would be happy to provide further methodological details or experimental data to further clarify these contributions.

Thank you very much for your valuable time and insightful comments on our manuscript. We have carefully considered each of your suggestions and have made the necessary revisions to improve the quality of our manuscript. Below, we provide a detailed point-by-point response to your comments, along with the specific changes made in the revised manuscript. Please see the attachment for the revised article.

Comments 1: -In line 39. The authors state that, “Firstly, it is difficult to generalize design rules”. Please elaborate further this argument, since there are many scholars writing about planting design guidelines and general rules exist.

Response 1: Thank you for your valuable suggestion. We mention that "it is difficult to generalize design rules" because many scholars have proposed guidelines for plant design. However, the factors involved, such as design concepts, climate conditions, and geographic environment, are diverse. This leads to higher learning costs and limits the applicability of these rules in different situations. As a result, design efficiency can be affected. Generative design can automate parts of the process. It reduces the time spent on manual adjustments and rule learning. This improves design efficiency and helps adapt more quickly to different design needs.

Comments 2: -Line 55-56.” There have been a small number of research results on the application of deep learning techniques in the field of landscape architecture, but there is a lack of research on small-scale plantscape design.” Please justify this argument with the appropriate bibliography.

Response 2: Thank you for your valuable suggestion. We have added appropriate references in the relevant sections to support this discussion.

Comments 3: -Avoid using the first singular or plural person in your sentence. Eg. Line 67. We implemented the study according to the following steps

Response 3：Thank you for your valuable suggestion. Based on your feedback, we have revised all sentences using the first singular or plural person to a more formal expression, such as "this study".

Comments 4: -In the introduction do not describe the methodology that you were following. Lines 67-79. The authors could briefly describe what the readers should expect to find in the following sections.

Response 4: Thank you for your valuable suggestion. We have revised the introduction section and removed the description of the methods. In lines 59-66, we have made the following changes: “This study presents the generation and comparison of small-scale plantscape designs using two algorithms, Pix2Pix and CycleGAN. The study begins by introducing the collection of flower border plans and the construction of a dataset based on scientific plant classification and labeling. It then describes the design process for generating plantscapes and compares the results generated by the two algorithms. Differences between the results of different algorithms and those of manually designed plans are discussed. The study aims to provide new tools and perspectives for landscape design through GAN technology, offering innovative approaches for plant configuration and design.”

Comments 5: -The research questions to be at the end of Introduction before the paragraph with the goals.

Response 5: Thank you for your valuable suggestion. We have made the changes based on your feedback. The research question has been moved to the end of the introduction (lines 150-161).

Comments 6: -From the section materials and methods, please extract any results and add them in the separate section of “Results”.

Response 6: Thank you for your valuable suggestion. As you suggested, we have moved the content related to the results from the Materials and Methods section to the Results section, specifically under 3.1. Two Rounds of Test Results and 3.3. Optimization of the Results . This helps to clearly separate the research methods and results.

Comments 7: -The same applies to the “Discussion”. There are a lot of examples that could be added to the Results section. Tables or graphs from the Discussion section should be included in the Results section. Please try to organize your results in the same structure of paragraphs as presented in the Discussion section. For example: 4.1.1. Image Quality, 4.1.2. Design Mode, etc

Response 7: Thank you for your valuable suggestion. Based on your feedback, we have moved the charts and results mentioned in the Discussion section to the Results section. We have also reorganized the structure of the Results section to align with the Discussion section. Following your suggestion, we have revised the relevant sections to 3.4.1. Image Quality and 3.4.2. Design Mode, etc., to present the experimental results more clearly. In the Discussion section, we need to evaluate the generated results of the better-performing CycleGAN model. This includes both an evaluation of image quality and an assessment of the design. In order to preserve the integrity and completeness of this section, we regret that we are unable to move the tables or graphs presented here to the Results section.

Best regards.

We believe that the revisions have significantly enhanced the clarity and quality of our manuscript. We have highlighted the changes in the revised manuscript for your convenience.

Thank you once again for your support and guidance. We look forward to your positive response.

Sincerely,

Sun Yuting

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

Thank you for your effort and all the improvements you have made. There is still room for more improvement and in further text, you will see my suggestions:

1. Although you have changed the title, the term “study” is still at the beginning of the title, which is technically not correct. I kindly ask the authors to consider the suggestion and adjust the title accordingly. One of the suggestions would be: “The applicability of two generative adversarial networks to generative plantscape design: a comparative study”.
2. Related to the aim of the research, thank you for accepting the comments, but I would suggest that you be more precise. How this algorithm will assist landscape architects is not clearly outlined in your goal. You cannot plan flower borders without considering the entire composition; it’s not just about the ecological aspect, but also the overall spatial characterization, each locality is specific. The aim of the paper, written like this, is general, so please be precise (lines 156-159). Did you use the text under the subsection “2.1. Analytical Framework”? You should.
3. Related to your response 7 it is clear a bit. Still, if I understand correctly, you want this work to represent a potential model/helper for landscape architects in selecting flower borders for projects implemented in Eastern China or not? If my observation aligns with yours, you still haven't convinced me that this model is necessary for landscape architects. Please strengthen your goals.
4. Please explain in more detail Figure 4 and Figure 7b.
5. The supplementary file is not very clear. It is not obvious which categories and parameters were used for the selection. Only species with Latin names are listed, along with the columns 'R', 'G', and 'B'. Are the numbers in these columns the numbers of individuals in red, green, and blue colour? Please clarify and explain what 56 subcategories mean. Also, please clarify the differences between the terms '56 subcategories' (line 235) and '56 elements' (line 247).
6. Lines between 361-370, what is the purpose of this sentence " ’Rounds of training’ refers to the number of times the model processes the dataset during training”. Still Table 3 and 4 are not clear what you want to achieve with them.
7. In response 22 you wrote: “Machine-generated results can be a starting point for design. However, these results often have flaws and cannot be directly used in practical design.” Two questions are opened. The first one is – what do you mean by a starting point? The second one is – if this model can not be used in practical design what is the purpose of its usage? The answer to the second question should be written in the aim of the research.
8. Regarding response 24. Namely, the study of 'Generative Adversarial Networks' is not limited to researchers from just one continent, it is international. It is very important that references are not only from one continental area; by doing so, you open pathways of thinking for scientists from other regions and continents to apply your methodology or something similar in their research. Therefore, the Discussion section should be oriented in this way, but the reference list must be broader, encompassing more than one continent. It does not matter whether the article is in Chinese, English, German, French, etc.; what matters is that you have taken into account research from other continental areas during your study. This way, you satisfy the scientific structure of the discussion, and you strengthen your research, making it more applicable. I kindly ask the authors to respond to this comment adequately.
9. The Discussion section still does not meet the scientific format. You do not need to discuss the results solely and exclusively within the scope of your own work. It is necessary to compare your results with others in order to discuss the entire problem you are addressing. Please see other MDPI articles so that you can gain insight into the structure of the Discussion section.
10. Please explain more informative Table 7. Please explain the transformation from “generated image” through “optimization result” to the “modelling pictures”.

 

Best regards.

Author Response

Dear Reviewer,

Thank you very much for your valuable time and insightful comments on our manuscript. We have carefully considered each of your suggestions and have made the necessary revisions to improve the quality of our manuscript. Below, we provide a detailed point-by-point response to your comments, along with the specific changes made in the revised manuscript.

Comments 1: Although you have changed the title, the term “study” is still at the beginning of the title, which is technically not correct. I kindly ask the authors to consider the suggestion and adjust the title accordingly. One of the suggestions would be: “The applicability of two generative adversarial networks to generative plantscape design: a comparative study”.

Response 1: Thank you very much for your valuable suggestion. Based on your recommendation, we will modify the title to: "The applicability of two generative adversarial networks to generative plantscape design: a comparative study"

 

Comments 2: Related to the aim of the research, thank you for accepting the comments, but I would suggest that you be more precise. How this algorithm will assist landscape architects is not clearly outlined in your goal. You cannot plan flower borders without considering the entire composition; it’s not just about the ecological aspect, but also the overall spatial characterization, each locality is specific. The aim of the paper, written like this, is general, so please be precise (lines 156-159). Did you use the text under the subsection “2.1. Analytical Framework”? You should.

Response 2: We sincerely appreciate your thoughtful suggestions. In response to your recommendation, we have made the research objectives more specific in lines 156-170. The updated research objectives now explicitly state that this study aims to compare the applicability of different GAN algorithms. We aim to identify which algorithm is better suited for generating plantscape design solutions that meet the spatial characteristics, aesthetic coherence, and ecological suitability of flower borders. Additionally, we have provided further clarification on the model’s usage. Designers need only to input the site boundary, enabling them to quickly obtain a preliminary plant layout plan in the early stages of design. This not only improves design efficiency but also supports the conceptualization process. To enhance the specificity of the research objectives, we have also referred to Section 2.1 and expanded on the dataset construction. The data is derived from high-quality plant configuration projects in East China, which share similar climate and soil conditions. This consistency helps the model learn region-specific design features.

The specific changes are as follows: "This study proposes an experimental procedure for a GAN-based plantscape plan generative design model. It focuses on comparing different GAN algorithms to create flower border plans that reflect spatial characteristics, aesthetic coherence, and ecological suitability. To ensure the validity and contextual relevance of the experiment, the dataset was constructed from high-quality plan drawings. These were collected from completed plant configuration projects in urban sites across East China, where similar climate and soil conditions provide a consistent ecological basis. The samples were selected for their representativeness and design diversity, enabling the model to identify and learn region-specific design features. The generative model is designed to integrate into existing workflows, aiding human-computer collaboration in early-stage plantscape design. By inputting site boundaries, designers can quickly obtain preliminary flower border layouts. This helps designers create initial concepts faster, reducing repetitive tasks and improving efficiency. The model supports, rather than replaces, the designer’s role, offering value in concept generation and early design decisions."

 

Comments 3: Related to your response 7 it is clear a bit. Still, if I understand correctly, you want this work to represent a potential model/helper for landscape architects in selecting flower borders for projects implemented in Eastern China or not? If my observation aligns with yours, you still haven't convinced me that this model is necessary for landscape architects. Please strengthen your goals.

Response 3: Thank you very much for your valuable comments. We truly appreciate that your understanding aligns with our research objectives. This study aims to introduce a GAN-based plantscape generation model to support landscape designers in creating flower border designs. In response to your concerns regarding the necessity of the model, we have provided further clarification in the background section to highlight its practical value. At present, plantscape and flower border design face several significant challenges in practice. First, the wide variety of plant species requires designers to consider numerous factors, often relying on their experience and intuition. Second, the initial stages of design present a high threshold, requiring designers to possess both extensive plant knowledge and strong aesthetic skills. Third, plant arrangements are often highly interdependent, leading to low efficiency in making design adjustments. Finally, the design process involves a substantial amount of repetitive tasks, which offer considerable potential for automation.

In light of these challenges, the model proposed in this study is designed to provide designers with preliminary plant arrangement suggestions based on site boundary conditions during the early stages of design. This tool not only serves as an inspirational aid for composition and plant selection but also reduces repetitive tasks, significantly improving efficiency, especially in time-sensitive or resource-constrained projects. By integrating deep learning with landscape design, this research seeks to contribute to the digitalization and automation of plantscape design, thereby demonstrating the model's practical applicability and potential for future development.

In response to your recommendation, we have made the necessary revisions and emphasized these points in the research objectives section of the paper (lines156-170): "This study proposes an experimental procedure for a GAN-based plantscape plan generative design model. It focuses on comparing different GAN algorithms to create flower border plans that reflect spatial characteristics, aesthetic coherence, and ecological suitability. To ensure the validity and contextual relevance of the experiment, the dataset was constructed from high-quality plan drawings. These were collected from completed plant configuration projects in urban sites across East China, where similar climate and soil conditions provide a consistent ecological basis. The samples were selected for their representativeness and design diversity, enabling the model to identify and learn region-specific design features. The generative model is designed to integrate into existing workflows, aiding human-computer collaboration in early-stage plantscape design. By inputting site boundaries, designers can quickly obtain preliminary flower border layouts. This helps designers create initial concepts faster, reducing repetitive tasks and improving efficiency. The model supports, rather than replaces, the designer’s role, offering value in concept generation and early design decisions."

 

Comments 4: Please explain in more detail Figure 4 and Figure 7b.

Response 4: Thank you for your valuable suggestions, and we apologize for any confusion caused by the article. In response to your comment regarding Figure 4, we have added further detailed explanations to the original text. The additional details are as follows:

Lines 213-226 provide a more detailed explanation for Figure 4: Construction Drawings for Flower Borders.

Lines 226-228 offer a detailed explanation for Figure 4: Plant Statistics.

Lines 229-253 present the detailed explanation of the classification standards and principles in Figure 4: Screening.

Lines 259-265 describe the detailed explanation of the color selection method in Figure 4: Annotating.

Lines 265-268 explain Figure 4: Batch Processing in more detail.

Lines 276-281 provide further details on Figure 4: Image Enhancement.

Additionally, in lines 421-425, we provided a more specific explanation of Figure 7b. "Figure 7b shows the loss function of the generative and discriminative networks during the training of the CycleGAN model. As the training iterations accumulate, the loss functions of the respective generators and discriminators gradually converge and occlude together, indicating that the training of the neural network is close to being perfect." This image illustrates the convergence of the two loss functions, signifying that the CycleGAN model training is effective.

 

Comments 5: The supplementary file is not very clear. It is not obvious which categories and parameters were used for the selection. Only species with Latin names are listed, along with the columns 'R', 'G', and 'B'. Are the numbers in these columns the numbers of individuals in red, green, and blue colour? Please clarify and explain what 56 subcategories mean. Also, please clarify the differences between the terms '56 subcategories' (line 235) and '56 elements' (line 247).

Response 5: We sincerely apologize for the confusion regarding this issue. In the supplementary file, the 56 species refer to the 56 plant elements selected for the study. These include broader plant categories such as trees and styled shrubs, as well as more specific plant categories, including those identified by their Latin names. The classification and selection of these plants were primarily based on their frequency of use within the dataset. In response to your comment, we have provided a more detailed explanation of the classification principles in lines 229-234.

In annotating, the color model used is RGB, with the columns labeled 'R', 'G', and 'B' representing the red, green, and blue channel values, respectively. The terms "56 subcategories" (line 246) and "56 elements" (line 259) both refer to the 56 species in the supplementary file. We apologize for any confusion caused by the article. In the revised manuscript, we have added clarifications and corrections (lines 241-244, lines 252-253).

 

Comments 6: Lines between 361-370, what is the purpose of this sentence " ’Rounds of training’ refers to the number of times the model processes the dataset during training”. Still Table 3 and 4 are not clear what you want to achieve with them.

Response 6: We sincerely apologize for not providing a clearer explanation in the text. The term " 'Rounds of training' refers to the number of times the model processes the dataset during training" is the explanation provided for the 'Epoch' in Tables 3 and 4. Tables 3 and 4 display the generated results for different site types at various training iterations for Pix2Pix and CycleGAN, respectively. The purpose of these tables is to visually show how the model's training improves as the number of training iterations increases. We believe that Tables 3 and 4 complement the loss functions shown in Figure 7, providing further validation of the model's progress.

In response to your concerns and to further clarify the content, we have added a more detailed explanation of the images in Tables 3 and 4 in lines 385-398.

It can be observed that the edges of the plaques generated by the model are gradually clearer with the increase of the number of iterations, and the image completeness is gradually improved. After 300 iterations, color patches gradually appear in the generated image and the results become relatively stable. As the training reaches 500 times, the output of the model is closer to the real image, the color block segmentation is gradually complete from dispersed, the division is clearer, and the scale and number of plant patches are reasonable, presenting a natural-style distribution. As shown in the figure (Table 3 and Table 4), with the increase in training rounds, the model's training gradually improves.

Considering these results, the study finally chose to set the number of training iterations to 500 for all of them. The samples in Table 4 represent different curved site shapes, including crescent, near-circular, and elongated forms. These variations are used to evaluate the model’s performance and adaptability in generating designs for different spatial layouts.

 

Comments 7: In response 22 you wrote: “Machine-generated results can be a starting point for design. However, these results often have flaws and cannot be directly used in practical design.” Two questions are opened. The first one is – what do you mean by a starting point? The second one is – if this model can not be used in practical design what is the purpose of its usage? The answer to the second question should be written in the aim of the research.

Response 7: 

Thank you very much for your thoughtful questions. Regarding the term “starting point” in the design process, we meant that the model-generated results can serve as a reference in the early stages of design. These results help designers quickly establish initial spatial structures and plant arrangements. This process saves time and inspires new ideas, allowing for further adjustments and optimization.

Regarding the second issue, we recognize its importance. Therefore, we have added further clarification to the research objectives section to better explain the model's role. While the generated results cannot be directly used for construction or final outputs, they provide an efficient tool to assist designers in the early stages of design. The model is not meant to replace designers but to improve efficiency, reduce repetitive tasks, and offer new methodological support for regional plantscape design. We have added the following explanation in lines 156-170: This study proposes an experimental procedure for a GAN-based plantscape plan generative design model. It focuses on comparing different GAN algorithms to create flower border plans that reflect spatial characteristics, aesthetic coherence, and ecological suitability. To ensure the validity and contextual relevance of the experiment, the dataset was constructed from high-quality plan drawings. These were collected from completed plant configuration projects in urban sites across East China, where similar climate and soil conditions provide a consistent ecological basis. The samples were selected for their representativeness and design diversity, enabling the model to identify and learn region-specific design features. The generative model is designed to integrate into existing workflows, aiding human-computer collaboration in early-stage plantscape design. By inputting site boundaries, designers can quickly obtain preliminary flower border layouts. This helps designers create initial concepts faster, reducing repetitive tasks and improving efficiency. The model supports, rather than replaces, the designer’s role, offering value in concept generation and early design decisions.

 

Comments 8: Regarding response 24. Namely, the study of 'Generative Adversarial Networks' is not limited to researchers from just one continent, it is international. It is very important that references are not only from one continental area; by doing so, you open pathways of thinking for scientists from other regions and continents to apply your methodology or something similar in their research. Therefore, the Discussion section should be oriented in this way, but the reference list must be broader, encompassing more than one continent. It does not matter whether the article is in Chinese, English, German, French, etc.; what matters is that you have taken into account research from other continental areas during your study. This way, you satisfy the scientific structure of the discussion, and you strengthen your research, making it more applicable. I kindly ask the authors to respond to this comment adequately.

Response 8: Thank you for your suggestion. Based on your recommendation, we have included research from other regions in the revised version to ensure that our study is more broadly applicable. The specific references include studies from China, Thailand, the United States, Iran, Canada, Germany, South Korea, Ethiopia, and Australia. The newly added references are highlighted in the Reference section.

 

Comments 9: The Discussion section still does not meet the scientific format. You do not need to discuss the results solely and exclusively within the scope of your own work. It is necessary to compare your results with others in order to discuss the entire problem you are addressing. Please see other MDPI articles so that you can gain insight into  the structure of the Discussion section.

Response 9: We sincerely appreciate your valuable suggestion. In response to your feedback, we have reviewed other MDPI articles and revised the discussion section accordingly. In the updated version (lines 532-558), we have included comparisons with other relevant studies. This allows us to highlight the similarities and differences between our findings and existing research, while also clarifying the position of our study within the broader context. By doing so, we hope to better demonstrate the contributions and innovations of our work and strengthen its connection to other research.We are truly grateful for your guidance. Your suggestion has been instrumental in helping us improve the structure of the paper.

This paper proposes a flower border layout generation system based on GAN algorithms. The system achieves the "generation, optimization, and validation" of complex small-scale plantscape designs. Built on Pix2Pix and CycleGAN, the system automatically generates flower border layout plans that meet spatial, aesthetic, and ecological requirements, providing landscape designers with an efficient tool and promoting the intelligent and scientific development of plantscape design.

In recent years, GAN has been widely applied in architecture and landscape design, particularly in the design and image generation of outdoor spaces such as parks and communities. These applications are typically focused on large-scale, functional areas with regular layouts. For example, Liu et al. employed generative design methods to study structured spaces like private gardens in Jiangnan, while Qu et al. developed a functional plan generative design method for residential landscapes based on CGAN. However, the application of GAN in small-scale plantscape design, particularly flower border design, remains relatively underexplored.

Recently, some studies have explored the use of AI models to generate flower border designs. For instance, Cui et al. used diffusion models to generate visual effects for flower borders. However, these studies largely focus on aesthetic outcomes, particularly for facades, and rarely address the spatial planning of plant layouts. In contrast, this research focuses on generating plant layout plans, not only considering spatial and aesthetic effects but also integrating scientific evaluation metrics. This approach ensures that the generated designs are visually satisfying while meeting spatial, ecological, and functional requirements, addressing the unique challenges of small-scale plantscape design.

By extending the application of GAN from large-scale urban planning to more detailed plantscape elements, this study introduces a new research methodology for plantscape design, contributing to a deeper understanding of how GAN can assist in designing small-scale landscapes.

 

Comments 10: Please explain more informative Table 7. Please explain the transformation from “generated image” through “optimization result” to the “modelling pictures”.

Response 10: Thank you for your suggestion. The process from "generated image" to "optimization result" follows the steps outlined in Section 3.3, "Optimization of the Results." From "optimization result" to "modeling pictures," the plants were modeled based on the optimized plan using Lumion, and the rendered images were then produced from the model. In response to your comment, we have added the following explanation in lines612-615：The generated images were first refined according to the steps described in Section 3.3 to produce the optimization results. These were then imported into Lumion for 3D modelling and rendering, resulting in the final modelling pictures.

We believe that the revisions have significantly enhanced the clarity and quality of our manuscript. We have highlighted the changes in the revised manuscript for your convenience.

Thank you once again for your support and guidance. We look forward to your positive response.

Sincerely,

Yuting Sun

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors

Thank you for your feedback to my comments. Your manuscript has been improved.

 

Author Response

Dear Reviewer,
We sincerely appreciate your time and effort in reviewing our paper. Your valuable feedback has greatly helped us refine our work. We are pleased to hear that you recognize our research and find it ready for publication.
Thank you again for your professionalism and support. We look forward to contributing to the field.

Best regards,
Yuting Sun

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

I would like to sincerely thank you for the effort you’ve put into improving the paper based on my corrections. Although I am not fully convinced about the applicability of your results and the model you are proposing, this does not mean that I am opposed to having these results published. However, before that, I believe it would be beneficial if you considered some comments to make your writing clearer and more understandable.

1. First of all, if the terminology throughout the paper is not consistent, it can lead to further confusion while reading. In Comment 5, I kindly asked you to clarify the differences between the terms "56 subcategories" (line 235) and "56 elements" (line 247). Instead, you introduced additional confusion in terminology by explaining that: "the 56 species refer to the 56 plant elements selected for the study" (your response to comment 5); and then you used various terms such as: “56 plant elements” (line 717), “56 subcategories” (246, 259), “56 valid colours” (line 403), “56 RGB colour values” (line 439), “56 marker colours” (line 441). Do all of these refer to the same concept? Please make the necessary corrections and use a single, consistent term for all of them.
2. Regarding the discussion, I kindly ask you to consider the comment from the previous review. I will repeat it here: "The Discussion section still does not meet the scientific format. There is no need to discuss the results solely and exclusively within the scope of your own work. It is important to compare your results with others in order to address the entire problem you are discussing. Please refer to other MDPI articles to gain insight into the structure of the Discussion section."

Therefore, the section you have added from lines 526 to 702 is correct, and it reflects the model in which the discussion should be structured. However, everything following that seems confusing, as it is still just solely a discussion of your results. I kindly suggest incorporating those comments into the Results section. This way, you can explain and comment on the results in parallel.

Additionally, it is not usual to include images in the Discussion section. Please consider this suggestion.

1. The reference list is still somewhat limited in terms of authors from outside Asia, which opens the question about the applicability of your results may not be fully global.

Best regards.

Author Response

Dear Reviewer,

We would like to express our sincere gratitude for the time and effort you have dedicated to reviewing our manuscript. Your constructive feedback is invaluable, and we deeply appreciate your thorough and thoughtful comments. We are fully aware of the importance of addressing the points you raised in order to enhance the clarity, quality, and scientific rigor of our paper. We have carefully considered each of your suggestions and have made revisions accordingly, which we hope will meet your expectations. Please find our detailed responses to your comments below.

 

Comment 1: First of all, if the terminology throughout the paper is not consistent, it can lead to further confusion while reading. In Comment 5, I kindly asked you to clarify the differences between the terms "56 subcategories" (line 235) and "56 elements" (line 247). Instead, you introduced additional confusion in terminology by explaining that: "the 56 species refer to the 56 plant elements selected for the study" (your response to comment 5); and then you used various terms such as: “56 plant elements” (line 717), “56 subcategories” (246, 259), “56 valid colours” (line 403), “56 RGB colour values” (line 439), “56 marker colours” (line 441). Do all of these refer to the same concept? Please make the necessary corrections and use a single, consistent term for all of them.

Response 1: Thank you very much for your valuable suggestions. We greatly appreciate your feedback and sincerely apologize for the inconsistency in our manuscript. In our paper, the terms "56 subcategories," "56 elements," and "56 plant elements" all refer to the same concept of “56 plant elements,” specifically focusing on the different plant types. We have now revised and unified these terms to "56 plant elements" to avoid confusion. On the other hand, the terms "56 valid colours," "56 RGB colour values," and "56 marker colours" refer to the colours associated with these 56 plant elements, emphasizing the concept of "color." To ensure consistency and clarity in our terminology, we have revised them to "56 valid colours". We sincerely hope that these changes will enhance the clarity and consistency of the manuscript.

 

Comment 2: Regarding the discussion, I kindly ask you to consider the comment from the previous review. I will repeat it here: "The Discussion section still does not meet the scientific format. There is no need to discuss the results solely and exclusively within the scope of your own work. It is important to compare your results with others in order to address the entire problem you are discussing. Please refer to other MDPI articles to gain insight into the structure of the Discussion section."

Therefore, the section you have added from lines 526 to 702 is correct, and it reflects the model in which the discussion should be structured. However, everything following that seems confusing, as it is still just solely a discussion of your results. I kindly suggest incorporating those comments into the Results section. This way, you can explain and comment on the results in parallel.

Additionally, it is not usual to include images in the Discussion section. Please consider this suggestion.

Response 2: Thank you very much for your valuable suggestions. Based on your feedback, we have carefully reassessed the structure of the Discussion section and made revisions accordingly. Following your guidance, we have moved the content from lines 526 to 702, including the images, to the Results section. This way, we can explain and comment on the results in parallel, as you suggested. Additionally, we have reviewed and referenced several relevant articles from MDPI journals to adjust and improve the Discussion section. We have ensured that it not only focuses on our own results but also compares them with existing studies. Furthermore, we have removed the images and made corresponding changes to the text (lines 675-722) to ensure that the Discussion section aligns with the formatting norms of academic journals. Once again, we deeply appreciate your suggestions, which have helped us improve the manuscript’s structure and rigor.

 

Comment 3: The reference list is still somewhat limited in terms of authors from outside Asia, which opens the question about the applicability of your results may not be fully global.

Response 3: Thank you very much for your valuable feedback on our reference list. We fully understand your concerns regarding the scope of the literature and sincerely appreciate your careful review. In our previous revisions, we made efforts to expand the reference list, particularly by searching for relevant studies globally. However, due to certain limitations in the availability of literature or language barriers, we may not have fully covered studies from regions outside Asia. We acknowledge this limitation and take it seriously.

In response to your concerns and to more rigorously reflect the limitations of our study, we have added the following statement in the Discussion section (Lines 713-718): “It is worth noting that the data used in this study primarily comes from the East China region, which may limit the direct applicability of the results to other geographic locations. However, we believe that the methodology can be adapted to different regions with the right adjustments to the datasets, allowing for broader application and comparison in future research.”

We sincerely hope that this addition makes our paper more thorough and better addresses your valuable comments. While the existing references are mostly concentrated on studies from Asia, we believe they still provide meaningful scientific value. We will continue to monitor developments in this field and expand the coverage of our references in future work. Thank you again for your constructive suggestions and support.

 

We hope these revisions address your concerns and improve the overall quality of the paper. We are confident that these changes will make the manuscript clearer, more comprehensive, and better suited for publication.

 

Best regards.

Author Response File: Author Response.pdf