Task-Aligned Oriented Object Detection in Remote Sensing Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

33-35: consider to add also medical applications

37:  arbitrary-oriented detection ->  arbitrary-oriented object detection

50: BBR is not explained

52: I would apprecitate some introductory sentence for the regression task. Some readers may be confused.

160: the printed figure is not readable. Consider to increase resolution and/or reorganizace the figure to enlarge text labels.

180-186: ReLU is usually used as an activation function. Nevertheless, some authors (e.g. Géron, 2019) recommend other functions with better behaviour such as Leaky ReLU or Exponential Linear Unit (ELU) or SELU.

283-284: It would be interesting to know the data shares for training, validation and testing.

306-354: “5.13% increase” and similar statements: consider that the appropriate unit name is “percentage points” and not real %. A percentage point or percent point is the unit for the arithmetic difference between two percentages.

306-354: “Consequently, the effectiveness of the X is affirmed.“ We do not know if 5.13 (2.39 etc.) perc.points are enough to distinguish random (natural) variability. The hypothesis of a significant difference should be rigorously statistically tested.

347, table 2: the headline label „MAP“ should be changed to “mAP“

365: „The method described in this paper obtain incorrect results during object detection“ – consider to change the sentence otherwise some readers may understand that all detected objects are incorrect.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors mention that the method known as YOLO is limited in its accuracy when detecting small objects. However, the most recent versions have demonstrated their effectiveness in this area. Were tests carried out with the most recent versions of YOLO and compared against the results obtained?

What percentage of the dataset was used for training and validation? It is suggested to explain this part in more detail and to provide justification for the selection of percentages.

The datasets used have different dimensions, and it is not mentioned which image resolution was considered during the training process.

Table 2 compares the methods presented in the state-of-the-art against its proposal. In general, the results obtained from the metrics are good, but not in all the proposed metrics. What are the reasons why your method does not present an overall advantage in all metrics? Were the results in Table 2 carried out by implementing all the state-of-the-art methods, or is it simply a numerical comparison that does not consider the same conditions?

Comments on the Quality of English Language

No comments about the English language.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors Summary: This work addressed an approach to solve the inconsistency problem between classification and regression and emphasized the problems with conventional convolutional operations.   Comments:  1. The introduction section should be addressed more comprehensively and also add recent works. 2. The manuscript is proficiently written. It is suggested to conduct a computational performance analysis of the proposed method for a better understanding of the outcomes. 3. Discussion part is short, It is suggested to discuss the impact of the proposed work and outcomes more comprehensively. 4. What do you think about the future work of the proposed method? It suggested to address the future improvement of the proposed method. Comments on the Quality of English Language

Minor editing is required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf