A Novel Hybrid Attention-Driven Multistream Hierarchical Graph Embedding Network for Remote Sensing Object Detection
Round 1
Reviewer 1 Report
1)Englishi should be improved, and many grammar errors must be corrected.
2)Fig. 2 should be modified, since it's too wide.
3) How do the hyper-parameters affect the performances, e.g., \delta in Eq. 1 and \eta in Eq. 2?
4)The experiments should be analyzed in more detail. For example, in Tab. 2 and Tab. 4, why performances of airplane and ship by the proposed approach are worse than MGCN?
5) Computation time of each approach should be compared.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Line 342: „In here, we note that the layer number of our proposed hierarchical graph is empirically set to c = 3“, Please elaborate on this. Can c be set to a higher number eg. C=4?
Line 407 and equation 26 „The overall multitask loss function“, please provide information about alpha and beta parameters. It continues on line 489:
„the two hyper-parameters α and β for the regularization in the overrall loss function are set to 0.01 and 0.05“,
…range for parameters {0.0001, 0.001, 0.005, 0.01, 0.005} may be too discreete?
Also, how do you explain different parameters for a different dataset?
Is L1 loss (line 410, eq 29) something proposed by authors, or can be found in literature? If this is novel, please elaborate on how did you acquire those formulas?
Line 486, To avoid the over486 fitting problem, we set the dropout rate with a probability of 0.5 on the set of experimental datasets.
Was this dropout rate too excessive? For hidden layers, the normal droput rate is up to 0.5, and up to 0.8 for input layers..
Please improve the quality of fig 6.
To avoid the overfitting problem, we set the dropout rate with a probability of 0.5 on the set of experimental
487 datasets. A??
It is common to provide ROC curves while presenting performance of models for object detection
And finally, the biggest comment: The paper lacks any information and comparison with other methods about training times and execution times, or complexity.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Globally, the manuscript is well written and very well organized. However, there are too many English errors that must be corrected. I recommend the authors to use a professional proofreading service or having the manuscript reviewed by a natural/mother tong speaking person. Please refer to the attached PDF file, where some of the needed corrections are highlighted.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors have answered all of my questions and the paper has been improved.
In my opinion it can be accepted for publication.
