Detection and Recognition of Spatial Non-Cooperative Objects Based on Improved YOLOX_L
Round 1
Reviewer 1 Report
In this article, the single-stage detector YOLOX_L was used as a baseline and the authors implement own methods on it to prove the effectiveness since YOLOX_L has a high performance. According to the various network structure functions, YOLOX_L can be divided into three parts, namely CSPDarknet, FPN and YoloHead. Why wasn't much detail given to CSPDarkNet, FPN and YoloHead? What is the difference between those and also the developed design? Why weren't the authors made state of the art and compared their design with other solutions? The literature on the topic of Detection and Recognition of Spatial very little in the paper. I see many scientific gaps in the paper that absolutely need to be addressed.
Author Response
I'm sorry to reply to your message at this time. As the first author of this article, I took part in a lot of written tests and interviews in September, the golden period of campus recruitment in autumn, and ignored our paper revision work. Please forgive me. As for the language expression problems in the article, we will ask professional personnel to modify them.
The following is my reply to your suggestion.
First of all, we describe some problems of space target detection in line 40 to 66.
Secondly, we introduce CSPDarkNet, FPN and YoloHead in detail in line 219 to 259. At the end of the paper, we conduct a series of experiments to compare the accuracy of some algorithms and our algorithm.
The details are in the file I sent to you. Sorry again, looking forward to your reply.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
1. Figure 2 and 3 had chinese words inside, please fix it.
2. English quality must be imporved for better reading, for example, in line no 196: "Therefore, As shown in Figure 3, driven by YOLOF....", some violations of grammar rules made the reading with difficulty.
3. In Figure 1 and 4, there was a term, C5, what did that mean? The authors should add the related definitions nearly thre figures.
4. Equation 1 made me a little confused, why xi existed in both side of the equation sign? and "Where xi is feature map from CSPDarknet,..." meant the left or the right one? Besides, was there any relation between equation 1 and 2? it looked similar in description while xi had different definition.
5. Table 5 seemed missing the column, Model, to present which row belonged to your proposed model.
Author Response
I'm sorry to reply to your message at this time. As the first author of this article, I took part in a lot of written tests and interviews in September, the golden period of campus recruitment in autumn, and ignored our paper revision work. Please forgive me.
As for the language expression problems in the article, we will ask professional personnel to modify them. The following is my reply to your suggestion.
Firstly, we have removed the chinese characters in Fig. 2 and Fig. 3.
Secondly, English quality will be improved for professional personnel if the article don’t have big problems.
Thirdly, C5 is a feature map in YOLO series. We have add the related definitions in the article. Then, for Equation 1, it is corrected and shown in the paper.
Finally, Table 5 shows whether our model has feedback connection, namely CBAM module, so we think it is OK here. If it is not clearly stated, we will improve it later.
Author Response File: Author Response.pdf
Reviewer 3 Report
There are some chinese characters in Fig. 2 and Fig. 3. Please remove them.
In addition, I am wondering whether Fig. 2 is necessary to the article, since it only provided the structure of an existing model.
What's the motivation to use CBAM in Fig. 4? How about other attention modules like SENet?
The related work section is not comprehensive. Some works are relevant to object detection and should be included, like Cascaded parsing of human-object interaction recognition.
More details and statistics about the proposed dataset should be provided, including how the data are collected.
Author Response
I'm sorry to reply to your message at this time. As the first author of this article, I took part in a lot of written tests and interviews in September, the golden period of campus recruitment in autumn, and ignored our paper revision work. Please forgive me.
As for the language expression problems in the article, we will ask professional personnel to modify them. The following is my reply to your suggestion.
Firstly, we have removed the chinese characters in Fig. 2 and Fig. 3.
Secondly, we think the Fig. 2 is necessary to the article, because this article is based on the YOLOX_L improvement, it is necessary to introduce it.
Thirdly, for CBAM, we bridge the performance gap between single-in single-out structure and multi-in single-out structure from spatial attention mechanism and channel attention mechanism. The main function of the module is the weight assignment of each channel. Just like Attention, it helps the network learn the important feature information.
Lastly, we provide more information in related work section and more details and statistics about the proposed dataset.
The details are in the file I sent to you. Sorry again, looking forward to your reply.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Thank you to the authors for editing the paper. The paper has been edited and now is clearer.
Reviewer 2 Report
The revised and added parts make the reading more easily. The certain terms comes from YOLO variants are also emphasized in this revision. The authors may add more descriptions (or explanations) in conclusion part about why the improvement in the research indeed works.
