All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The authors have addressed all of the reviewers' comments. This manuscript is now ready for publication.
[# PeerJ Staff Note - this decision was reviewed and approved by Mehmet Cunkas, a PeerJ Section Editor covering this Section #]
The revised version is significantly improved all all point addressed. Motivations and research objectives are now clear.
- Implementation details are provided with clear instruction of hyperparameter tuning.
- Authors have conducted the ablation study as suggested. I appreciated author's effort in conducting additional experiments. It's fine if authors doesn't want to include these results into the main text. However, I recommend authors to add a citations for the statement indicating the dominance of ARAE over VAE and GAN.
- An additional metric has been computed and updated.
Validity of the findings now meet the criteria with statistical proof for model reproducibility.
None
The manuscript meets the journal's standards. No further improvements are needed.
Experimental design meets the journal's standards. No further comments.
No more comments.
No more comments. The manuscript can be accepted for publication.
Please address all the comments from the three reviewers and revise the manuscript accordingly.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
This work proposed a computational method based on adversarially regularized autoencoder architecture to produce Chinese landscape art. The topic is quite interesting with values contributing to the development of modernization of digital art for traditional paintings.
The background provides good understanding of the topic to all-level users with simple but professional English. The manuscript is well-organized with clear sections. Generally, the manuscript is well-prepared with logical flow of reading. Despite some merits, this work remains technical issues that need to be addressed.
The experiments are clearly defined but they are too simple. Additional experiments need to be conducted to improve the quality of this work.
(1) Have authors tuned the parameters of the proposed model? It's noted that the model should be tuned to obtained the best version for generation.
(2) Ablation study is required to assess the contribution of each component in models.
(3) Using FID only is inadequate for measuring the quality of the work. Other metrics, such as Inception Score (IS), Structural Similarity Index (SSIM), etc., need to be calculated to provide overall assessment.
(4) Motivations for choosing adversarially regularized autoencoder architecture is unclear. Please explain in more details why adversarially regularized autoencoder is more appropriate compared to other simpler generative model.
Minor points:
- Figure 3: Please check the x-axis. It should be "Ours" not "Our"
- Figure 4: What do you mean by "fusion"? Is it contention or something else?
No statistical evidence is found. Please provide proof of model reproducibility. Authors need to repeat the modelling experiments and record all metrics. Descriptive statistics information (e,g., mean, standard deviation, 95%CI, etc.) is required.
None
This paper presents a new deep learning model for generating Chinese landscape art images in a controllable way. The manuscript is well-structured and includes all the key sections in a clear and logical order. The Introduction is detailed and easy to read, with helpful subsections that make the main ideas clear. The dataset preparation process is well explained, covering important steps like resizing, cropping, and removing calligraphy to keep the data consistent and clean. The Methodology section is also clear, with well-written equations and explanations that show a solid understanding of the methods used. However, the paper could be improved by explaining why this particular model architecture was chosen and how it compares to other possible approaches for the same task.
The experimental design is explained fairly well, but some key details need to be clarified:
- The paper should clearly describe how the data was split (for example, the training/test ratio) and how many samples were included in each group.
- The authors should also mention the training time and the computing setup used, as this information helps readers understand how practical and efficient the model is.
- Although the current evaluation metrics are suitable, it would strengthen the study to include more image quality measures, such as the Inception Score (IS), which assesses both the diversity and visual realism of the generated images.
The authors should test the model several times independently to check whether it produces consistent results across different runs.
• The abstract effectively integrates art preservation and deep learning innovation, demonstrating clear objectives, a defined problem, technical novelty, and practical implications. The paper is well-written and structured; however, there are some errors in reporting.
• Authors should include appropriate citations to support their statements regarding the challenges of globalization and the decline of traditional practices in Chinese landscape painting. It is important for the authors to elaborate on the role of digital preservation and restoration, rather than merely stating the decline of traditional art practices.
• Check error in-text citation in line 90.
• Figures 1, need caption to support discussion.
• The authors should clarify what is novel in their model architecture or training strategy compared to prior works.
• As the study focuses on traditional Chinese art, the authors should provide a brief justification or reference that explains the underlying aesthetic principles or cultural essence being represented to reinforce the credibility of their cultural preservation argument.
• The integration of “customized neural architectures with domain-aware training strategies” is interesting but vaguely stated. A concise of how these strategies differ from existing approaches would help to grasp the technical innovation.
• The paper should include a clear justification for selecting Chinese landscape art as the focus of study, explaining its significance and relevance to the research objectives.
• Based on the samples chosen, the experimental design does not appear to capture the typologies of Chinese landscape art styles. Authors should justify the sampling criteria and explain how the dataset represents the range of artistic traditions.
• The rationale for the dataset size is not provided. The authors should clarify why 3,560 samples are considered sufficient for model training and evaluation.
• While the dataset description effectively outlines the structure and curation process, the dataset split (training, validation, testing) and corresponding proportions are not explained. Providing this information is essential to ensure transparency and reproducibility of the experimental setup.
• Several key experimental details including latent dimension, random seeds, training duration, model parameters, and baseline setup are missing. These details are necessary to validate the proposed deep learning model’s performance and support reproducibility.
• Though the model architecture is well explained, the authors should clarify how it captures the unique cultural and artistic features of Chinese landscape art. Without this explanation, the paper appears mainly technical and lacks cultural grounding.
• Although the use of a curated dataset and FID metric is appropriate for quantitative evaluation, assessing cultural authenticity in traditional Chinese landscape art requires qualitative input from experts such as historians, visual artists, or cultural scholars. Incorporating such expert evaluations would provide a more holistic and credible validation of the model’s cultural and artistic relevance.
• The results and discussion sections need improvement, particularly with stronger support and elaboration on the validation of cultural authenticity of Chinese Landscape painting.
• The proposed model demonstrates promising results, and the preservation of Chinese landscape art is adequately addressed in the results and findings section.
• Even though the model results are acceptable, the study lacks sufficient methodological details, such as dataset division, parameter settings, training conditions, and code availability.
• The conclusion states that this paper presents a comprehensive framework using deep generative modeling to create Chinese landscape paintings, focusing on both technical challenges and cultural meaning. However, the findings should be supported by a more detailed and well-developed framework combines various steps, tools, and techniques.
• The English is generally readable but requires improvement in formatting and consistency.
• The dataset’s copyright, licensing, and ethical considerations need to be clearly stated.
• Include more recent and relevant citations in the reference list to strengthen the literature support.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.