R2FA-Det: Delving into High-Quality Rotatable Boxes for Ship Detection in SAR Images
Round 1
Reviewer 1 Report
Comments formulated during my review are presented below:
1) The English language in the paper have to be edited for publication.
2) The author must cite more recent articles that make use of relevant ship detection techniques, as the papers reported in the following items:
I) Pan, Zhenru, and Rong Yang. "MSR2N: Multi-Stage Rotational Region Based Network for Arbitrary-Oriented Ship Detection in SAR Images.".
II) Xiao, Xiaowu, et al. "Ship Detection under Complex Backgrounds Based on Accurate Rotated Anchor Boxes from Paired Semantic Segmentation.".
III) Chen, Chen, et al. "MSARN: A Deep Neural Network Based on an Adaptive Recalibration Mechanism for Multiscale and Arbitrary-Oriented SAR Ship Detection.".
3) How many times was each network trained and for how many epochs?
4) You accurately described the use of the anchors, but just for curiosity I would like to understand what the processing is for the anchor with score between 0.4 and 0.5.
5) You used the Adam algorithm for the optimization, but you do not provide the assigned values for all the parameters, for example the exponential decay rate for the first moment estimates and the exponential decay rate for the second-moment estimates. Did you set these parameters to default values?
6) The authors need to present and discuss several solid future research directions.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper considers ship detection for SAR imagery.
The paper is well written with an easy to follow reading flow.
A minor comment is that for completeness regarding current solutions, MultiLayer Perceptrons (MLP) based neural networks should also be mentioned such as,
O. Kechagias-Stamatis, N. Aouf and D. Nam, "Multi-Modal Automatic Target Recognition for Anti-Ship Missiles with Imaging Infrared Capabilities," 2017 Sensor Signal Processing for Defence Conference (SSPD), London, 2017,
pp. 1-5, doi: 10.1109/SSPD.2017.8233244.
WANG, Juan; SUN, Lijie. Study on ship target detection and recognition in SAR imagery. In: 2009 First International Conference on Information Science and Engineering. IEEE, 2009. p. 1456-1459.
JUAN, Wang; LIJIE, Sun; XUELAN, Zhang. Study evolution of ship target detection and recognition in SAR imagery. In: Proceedings. The 2009 International Symposium on Information Processing (ISIP 2009). Academy Publisher, 2009. p. 147.
ZHU, Jiwei, et al. Projection shape template-based ship target recognition in TerraSAR-X images. IEEE Geoscience and Remote Sensing Letters, 2016, 14.2: 222-226.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript presented the uses of a Rotate Refined Feature Alignment Detector ( R2FA-Det) method for ship detection. They claim that the method achieved a better performance than some of the state of art methods (not all? Or context?) with optimized processing speed. The experiment was designed well and performed a comprehensive analysis and presented the result reasonably. Still, few issues should be addressed before going for the publications. The specific issues are as follows.
The manuscript is prepared well however, the content of the introduction part is a bit confusing and a part of it (line 88 to 105) may better suit in the discussion than in the introduction. Content in line 70 to 88 is recommended to improve and focus to clarify the objective of the study than for solution directly.
Several equations are neither cited properly nor explained clearly and the number of variables is not defined, the variables (operations according to the line right before the equation) are never defined or cited. Many other equations (equation 3 defines the softmax but never cited???) were also not cited or explained. I strongly recommend to explain the equation well and define the variables properly.
Numer of terminologies in the manuscript such as RBox (line 207) are never cited.
Equation 15, average precision may not be correct. I recommend checking the equation. If you think this is correct, please develop the full equation or cite it properly.
The authors mentioned using the number of datasets (Table 1) but they have not categorized the performance of the proposed and referenced methods in different datasets mainly wave-bands. It would be more interesting to see the results of the method with a different frequency of the SAR dataset instead of mixing all these datasets together.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have prepared the manuscript well and addressed all of my comments. It is now ready to publish.
