AF-OSD: An Anchor-Free Oriented Ship Detector Based on Multi-Scale Dense-Point Rotation Gaussian Heatmap

Round 1

Reviewer 1 Report

In this manuscript, an end-to-end anchor-free oriented ship detector is proposed, for which the multi-scale dense point rotation Gaussian heatmap (i.e., MDP-RGH) is designed to solve the sample imbalance problem and suppress the interference of negative samples. Moreover, a multi-task object size adaptive loss function is introduced to implement the training process. The experimental results on HRSC2016 and DOTA verify the effectiveness of the proposed method. The novelty is acceptable for REMOTE SENSING, however, I have the following comments:

1. Although the proposed method mainly focuses on ship target detection, what about the applications to other remote sensing targets? Please discuss in the text or in the conclusion.

2. In Section II, the deep study of the innovation of the method for defining the target label should be provided, and the advantages and disadvantages of the label with respect to other data labeling methods also should be investigated.

3. In Section V, please highlight the significance and purposefulness of the compared network models, and then some recently-proposed methods based on Gaussian heatmap should be added for the comparison to further verify the effectiveness of the proposed method.

4. Please more accurately compare the underlying parameter number of the comparative methods as well as the time complexity of the corresponding network models in Section V.

5. The equations in the manuscript contain more complex definition symbols that are easily misunderstood by the readers, which should be revised and streamlined.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The research are interesting as well as the proposed method.

Experiment design is well, so my remakrs relies mostly to description part.

 

 

- introduction - very general references are given in some places, like [1-7]. please be more precise

- lines 175-176 - it is not entirely clear - can you please provide an example?

- line 346 - how? did you introduce new examples or did you use augmentation techniques?

- Evaluation metrics given in 4.3 are not clear - please provide more details

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors  present a study  in  an end-to-end anchor-free  oriented ship detector (AF-OSD) framework based on a multi-scale Dense Point Rotation Gaussian Heatmap (MDP-RGH).  The study of airborne remote sensing scenes  with  strong background and noise  and  Interference  is a critical and challenging task in remote sensing. This is a hot topic in the   remote sensing área.   The presnt study  is well conducted and discussed.  It  proposes AF-OSD based on a MDP-RGH  for arbitrary-oriented ship detection. It is a nice piece of work.  Deserves publication as it is

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

In summary, the end-to-end anchor-free oriented ship detector (AF-OSD) framework is proposed to address the challenges of ship detection in remote sensing scenes. This framework is based on the multi-scale Dense Point Rotation Gaussian Heatmap (MDP-RGH) which models the oriented ships based on their shape and direction, while also adapting to the imbalance between positive and negative samples. Additionally, a multi-task Object Size Adaptive loss function is used to guide the training process and improve the detection quality and performance. Simulation results show that the proposed method significantly outperforms the state-of-the-art methods and demonstrates its effectiveness in detecting multi-scale oriented ships in remote sensing scenes. After carefully reviewing this manuscript, some major and minor concerns have been pointed out as follows:

Major comments:

1. The limitations of this work should be pointed out.

2. Not all symbols are well defined in the manuscript. In Eq. (25), what does N_c denote?

Minor comments:

1. One extra right parentheses in the first sentence, second paragraph, in Sec. 1.

2. Sec. 6  (Discussion) in the last paragraph in Sec. 1 is missing.

3. In Fig. 6, does Fig. 6(c) illustrate large scale representation for small ships or small scale? The corresponding Gaussian heatmap in Fig. 6(f) is small scale.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf