Improved YOLOv8 Model for Lightweight Pigeon Egg Detection

Round 1

Reviewer 1 Report

An effective and light-15 weight network model, YOLOv8-PG has been proposed and verfied, an interesting research has been made. In the manusript,

1. In the data qcquisition,  video data of pigeon eggs from April 1, 2022, to May 15, 2022. So why data before two years were used in the current research. And, only 150 paris data 

2. Figure 10, less quality, less informoation were presented in the figure. Figure 10 should be redrawed to show more details. and further discussion  should be included.

3. How the algrithom handle the RGB of the captured images.

4. Good job has been made, howere, little results has been discussed. Some further investigation on YOLOv8-PG network model can been added to insure the algratiom accuracy and efficency.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors deployed an object detection model, YOLOv8, with a modified architecture to reduce computing demand while gaining additional detection precision in the task of detecting pigeon eggs. The manuscript elaborates well on the practical problem of pigeon production and explicitly describes the methodology in model engineering. It also modestly guides the readers to acknowledge the limitations of the work. The manuscript is believed to be an excellent example for future studies in computer vision or relevant work. Hence, I strongly recommend this work be accepted after the minor revision can be completed accordingly.

Major comments:

-       Term consistency: The prediction classes, real and fake eggs, were not described consistently over the entire manuscript. For example, in Line 50, the eggs to replace the real eggs were called “fake eggs,” In Line 126, Caption in Figures 8 and 10, they were called “false eggs.” Since “True” and “False” are specific terms in classification problems. Using “False” without referring to a wrong prediction in such a task may introduce confusion to readers. Hence, I suggest using “fake eggs” to minimize the ambiguity.

-       Lack of details in Table and Figure captions: Although the authors provide abundant information, it successfully makes the work more convincing. Many of the figures and tables lack essential information to be self-explanatory. I have pointed them out specifically in my minor comments.

-       Abbreviation: Many letter abbreviations were presented without explaining their first occurrence.

-       Educational standpoints: Since it is an agriculture/biology journal, it is necessary to describe more of the concepts that might be recognized as common sense in the community of computer vision. For example, what are the definitions and functions of the “Input,” “Backbone,” “Neck,” and “Detection Head” network? Also, would it be beneficial if the authors could explain the merit of C2f (YOLOv8) over C3 (YOLOv5) before replacing it with the proposed architecture (YOLOV8-PG)?

Minor comments:

-       Line 178: Define C2f.

-       Line 186: Explain k (kernel size?) and ‘cp’.

-       Line 189: Explain t.

-       Figure 4: No sections (a) and (b) were found in the figure. Also, “h,” “w,” “k,” “cp,” “BN” (batch normalization?), and ReLU (not Relu) should be briefly explained in the figure caption. It should be similar to what they were presented in lines 213-215.

-       Figure 5: Same to the figure 4. No sections (a) and (b) were found in the figure.

-       Figure 6: Explain terms including “gs^2”, “sh,” “sw,” “g,” and “o.”

-       Figure 7: Great figure. It can be more helpful to the reader if the author can present the full names of “CF, “CFE, “SPPF, and “CBS” in the caption.

-       Table 2.: “Ir0” should be “lr0” (initial learning rate). The first letter is not correct.

-       Table 2.: Including the hyperparameters in the warm-up stage (the first 3 epochs) would be better.

-       Table 3.: The modules A to D should be described in the table caption, not only in the manuscript (Lines 333 to 338).

-       Table 3.: What is the confidence threshold for the F1 metric?

-       Line 371: The space should be moved to after the comma mark: “For the more stringent ,”  -> “For the more stringent, “

-       Figure 9. The meaning of the colors (Line 397) should also go to the figure caption.

-       Figure 10. True and false should be real and fake (eggs). Again, using “true” and “false” in a multi-classification problem can be confusing.

-       Line 399: Reference for Grad-CAM?

-       Line 429. Should be real and fake pigeon eggs.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf