Next Article in Journal
Influencing Factors, Kinetics, and Pathways of Pesticide Degradation by Chlorine Dioxide and Ozone: A Comparative Review
Previous Article in Journal
Influence of Agronomic Practices on the Antioxidant Activity of Three Mediterranean Officinal Wild Plants: Silybum marianum, Achillea millefolium, and Trifolium pratense
 
 
Article
Peer-Review Record

Panoptic Segmentation Method Based on Feature Fusion and Edge Guidance

Appl. Sci. 2025, 15(9), 5152; https://doi.org/10.3390/app15095152
by Lanshi Yang 1, Shiguo Wang 2 and Shuhua Teng 1,*
Reviewer 1:
Reviewer 2:
Reviewer 3: Anonymous
Reviewer 4:
Appl. Sci. 2025, 15(9), 5152; https://doi.org/10.3390/app15095152
Submission received: 21 February 2025 / Revised: 2 April 2025 / Accepted: 8 April 2025 / Published: 6 May 2025

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

- Include relate works section.

- Include and describe paper contributions in separate sections.

- Describe 3.1 section .

- Include LOE, SSIM,PSNR, FID metric achievements of your approach with original segmentation label images based on Figure 5, and compare with other research works.

- What is the limitation of your research work?

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors, great thanks for such a voluminous research of the proposed panoptic segmentation method based on feature fusion and edge guidance (PSM-FFEG). The paper contains enough scientific information about the network infrastructure, layers, and modules. The extensive experiments on the Cityscapes and MS COCO datasets as well as the analysis of the experiments results are presented in paper. The implementation details are also provided.

The main question addressed by the research is to improve the global field of computer vision using feature fusion and edge guidance with improvements proposed by authors in their developed framework based on the panoptic segmentation method.

The research is relevant to the field and addresses the specific gaps like multi-scale feature fusion limitations by the model’s adaptability to scene diversity when using the static nature.

The authors firstly proposed the panoptic segmentation framework, which includes significant innovations compared to the published studies to address the challenges of multi-scale feature fusion and boundary segmentation in the different complex scenarios.

The methodology is well described and does not require any improvements.

Conclusions contain descriptions of the complex scenarios used during experiments based on the details in tables with data from the main part of the paper.   The references are appropriate   Figures well describe the proposed method and layers/modules of architecture

So, good luck with resolving the future research endeavors and posting your source code to share your science with the scientific community.

Author Response

Thank you sincerely for your positive comments on our paper "Panoptic Segmentation

Method Based on Feature Fusion and Edge Guidance (PSM-FFEG)."

We are particularly grateful for your recognition of our contributions in addressing multi-scale feature

fusion limitations and boundary segmentation challenges in complex scenarios. Your acknowledgment

of our comprehensive methodology, extensive experimentation on Cityscapes and MS COCO datasets,

and thorough analysis is deeply encouraging to our team.

We are honored by your recommendation for acceptance and deeply appreciate your support of our

research. Your confidence in our work motivates us to maintain the highest standards of scientific rigor

in our future endeavors. We are committed to sharing our source code with the scientific community to

foster collaborative advancement in the field of computer vision.

Thank you again for your time and expertise. Your insights have been invaluable in refining our research,

and we look forward to continuing our work with the same dedication and thoroughness that you have

recognized.

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript titled "Panoptic Segmentation Method Based on Feature Fusion and Edge Guidance" proposes the PSM-FFEG framework to address current challenges in panoptic segmentation, specifically in multi-scale feature fusion and edge boundary definition. The authors introduce three innovations: (1) a dynamic multi-scale feature fusion mechanism leveraging deformable convolution and attention mechanisms, (2) an edge guidance module for enhancing boundary segmentation accuracy, and (3) a dual-path Transformer decoder for optimized global and local feature interactions. Experimental validations on the Cityscapes and MS COCO datasets demonstrate superior performance compared to other approaches, especially improvements in identifying small-scale objects.

- Adding more explanatory captions to figures and tables would help readers to understand the results more easily without having to refer back to the main text. Adding annotations that explain each component’s function and contribution to the overall architecture would enhance clarity.

- The statement about data availability should be provided at the end of the article.

- Sharing the source code for the proposed network, along with training data, scripts and pre-trained models, would significantly increase reproducibility.

- The authors could provide qualitative results and a more in-depth discussion of the limitations of the proposed method.

- I recommended to clarify and justify the technical choices regarding kernel sizes, dilations, and specific attention mechanisms.

- Add statistical analyses of variability across multiple training runs to reinforce the reliability of reported improvements.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The paper presents a novel panoptic segmentation method that integrates feature fusion and edge guidance to improve segmentation performance in complex scenes. While the proposed method builds on existing panoptic segmentation models, it introduces key innovations such as a dynamic multi-scale feature fusion mechanism, an edge guidance module, and a dual-path Transformer decoder. To improve the quality of the paper and ensure its suitability for publication, I recommend addressing the following remarks.

  • The introduction provides useful background information on panoptic segmentation but does not clearly explain how the proposed method differs from existing approaches. The problem statement should explicitly state what limitations of Mask2Former and other models the PSM-FFEG framework addresses.
  • The bibliography is weak. The authors should discuss object detection techniques, starting with classical methods (e.g., thresholding, mean-shift) and progressing to deep learning models (e.g., YOLO, Faster R-CNN, RetinaNet). This will help introduce the proposed approach by highlighting the limitations of previous methods.
  • The introduction should conclude with an overview of the paper's structure, summarizing the contents of the following sections.
  • The paper contains complex sentence structures that reduce readability. For example: "To tackle the limitations of existing methods, such as inadequate multi-scale feature fusion and ambiguous target boundary segmentation in complex scenes, we propose a novel panoptic segmentation network, PSM-FFEG (Panoptic Segmentation Model with Feature Fusion and Edge Guidance)." This sentence should be restructured for better clarity and readability.
  • The term "superior performance" is used without justification. The claims should be supported by statistical significance tests or clear numerical results with references.
  • The feature fusion mechanism lacks details. How does the model determine which features to retain or discard dynamically?
  • The edge guidance module is claimed to improve boundary segmentation. How does it compare quantitatively to traditional edge-aware segmentation methods?
  • The dual-path Transformer decoder is said to enhance feature interaction. Were ablation studies conducted to isolate and measure its impact?
  • The results section states that PSM-FFEG improves PQ scores over Mask2Former. Are these improvements statistically significant? Was a significance test (e.g., t-test) conducted to ensure that the improvements are not due to random variations?
  • The model outperforms Mask2Former by 2.6% on Cityscapes. What is the margin of error in these results? Were multiple runs conducted to verify reproducibility?
  • The method shows better segmentation for small objects (e.g., traffic lights, motorcycles). Which specific architectural components contributed most to this improvement?
  • Table 1 & Table 2: The performance differences between models should include standard deviations across multiple runs. Were all models trained under the same conditions and preprocessing steps?
  • Figure 1: The architecture diagram is not clearly labeled. Key components, such as the dual-path decoder, should be highlighted.
  • Figure 5: The segmentation visualization should include confidence scores or uncertainty maps to indicate where the model struggles.
  • How does the computational cost of PSM-FFEG compare to Mask2Former? Does the additional complexity significantly increase inference time?
  • Why was ResNet-50 selected as the backbone? Would a more advanced backbone (e.g., Swin Transformer) improve results?
  • How does the model handle occlusions and overlapping objects compared to previous methods?
  • What are the failure cases of PSM-FFEG? Are there scenarios where it performs worse than existing methods?

Finally, the paper presents a promising approach to panoptic segmentation with meaningful improvements in feature fusion and boundary refinement. However, several claims require further justification through statistical validation and comparative analysis. The clarity of figures and tables should be improved, and additional experiments are needed to verify the robustness of the proposed method. Addressing these points will strengthen the manuscript and enhance its scientific contribution.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Well edited

Reviewer 4 Report

Comments and Suggestions for Authors

I thank the author for considering my remarks. I recommend the publication of the paper in its current format.

Back to TopTop