Pairwise Guided Multilayer Cross-Fusion Network for Bird Image Recognition

Round 1

Reviewer 1 Report

This work presents a Progressive Cross-union Network that can recognize bird categories at a more specialized level. My main general comments are as below: 

- An important shortcoming is that the author does not highlight the contribution of their manuscript in comparison to the work that has been performed by previous researchers. This can be added in the introduction and/or conclusion section.

- The authors didn’t provide a comparison of the performances on training and testing sets. The authors should investigate experimentally the overfitting of the proposed model.

- The work will be significant if the source codes are presented to the public for a detailed analysis of the proposed method.

- Conclusions need more elaboration about: outcomes, limitations, and possible/future scenarios.

- The authors should investigate the stability of the proposed model because image can be degraded by additive noise, in the presence of cluttering backgrounds, geometric modifications such as pose changing and scaling, nonuniform illumination, and eventual object occlusions. Accept after minor revision.

Author Response

We feel great thanks for your professional review work on our article. As you are concerned, there are several problems that need to be addressed. According to your nice suggestions, we have made extensive corrections to our previous draft. Point-by-point response letter is in the attechment file.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors proposed a cross-layer information exchange and pairwise learning termed as "Progressive Cross-union Network (PC-Net)". After a thorough review, I lean towards a Major revision before any further proceedings. The reasons/suggestions are as follows:

1. The structuring of the manuscript from a reader's perspective is very sore on the eye (only in the first section) as it combined both Introduction and Related Works into one single huge chapter. This overwhelms the reader with numerous technical details in the first glance.

(a). I recommend the authors to separate the Section-I into Introduction and Related Works

(b). Also, I recommend the authors to add a Related Works Table that outlines the Research Work, Dataset used, Key aspects, Limitations. So, readers can actually find it interesting to draw parallels to the current research work alongside existing ones.

2. Add the paper organization paragraph at the end of the Introduction section which helps the reader to understand the section-wise flow of the manuscript. Once it is done, line 121 in the current manuscript becomes redundant.

3. Refine the equation one "\theta" as hyperparameter being used two times, mathematically, this could be two instances. However, it is recommended to use "\theta1, \theta2" if that makes sense for better understanding.

4. The proposed framework aspects such as CLFM, DSLM heavily relies on the loss functions CELoss and SGLoss. However, authors haven't presented any ablation study on how they chose these loss functions over the other alternatives. An ablation study regarding this must be presented in section 4.4 of the current manuscript regarding how the proposed framework aspects and selected components contribute to the overall accuracy.

5. Figure 8 caption must be replaced with the appropriate one

6. A part from the accuracy, the metric evaluations related to processing time and memory requirements (CPU FLOPS) must be added for better validation of the proposed method.

7. Finally, I would recommend the authors to add a List of Abbreviations after Conclusion for a better sum up of all the acronyms used along with the mathematical variables used in the manuscript at one place. 

Overall, I encourage the authors to modify the manuscript with these comments in mind and also fix the English in the Sections 2 and 3 of the current manuscript.

Authors must run an English-correction check through the Sections 2,3 and fix syntax and grammatical errors.

Author Response

We feel great thanks for your professional review work on our article. As you are concerned, there are several problems that need to be addressed. According to your nice suggestions, we have made extensive corrections to our previous draft. Point-by-point response letter is in the attechment file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have addressed all the elements mentioned in the major revision comments. 

Extra set of evaluations and restructuring of the manuscript is done as needed.

I would recommend this manuscript for publication in the esteemed journal.