Spatial–Spectral Cross-Correlation Embedded Dual-Transfer Network for Object Tracking Using Hyperspectral Videos
Round 1
Reviewer 1 Report
The authors presented the detail explanation, performance evaluation and much experiments about the proposed approach.
Therefore, encourage to publish this paper after checking the English language and style.
Author Response
Thank you very much for handling our manuscript remotesensing-1750257 and for giving us thoughtful and constructive comments and suggestions. Please see the attachment for responses to the comments.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors presented a hyperspectral video tracker by modifying Siamese network by using transfer learning, weight-based hyperspectral dimension, and spatial-spectral cross correlaton embedding.
The idea is very good, and followings should be considered for better paper.
1) 2.1 Transfer learning -> remove, it is already well-knonw technique
Instead, instert hyperspectral video technologies.
2) 3. Proposed method
- insert the proposed system structure, then explain each block clearly.
3) eq. (9): why did't you use training to get the weight? You just use image gradient information and eigen values.
Author Response
Thank you very much for handling our manuscript remotesensing-1750257 and for giving us thoughtful and constructive comments and suggestions. Please see the attachment for responses to the comments.
Author Response File: Author Response.pdf
Reviewer 3 Report
Novelty is questionable.
In compared to RGB footage, how does your technique perform on grayscale?
I'd recommend playing with the frequency domain to pinpoint the object in multiple frames or demonstrating how the algorithm handles this.
How were the training samples shifted, and how was multichannel data used?
Have you looked at using dense sampling to track the objects?
neural
Why do you require pooling in your strategy?
In the mathematical formuale, I am unable to locate your own contribution ot the method defined. It appears to be excessively generic. It's going to need a complete rewrite and focus only on the explanation of your method.
Experimental part has no rigorous statistical analysis nor proofs on overfitting.
Author Response
Thank you very much for handling our manuscript remotesensing-1750257 and for giving us thoughtful and constructive comments and suggestions. Please see the attachment for responses to the comments.
Author Response File: Author Response.pdf
Reviewer 4 Report
1. SiamRPN++ and SiamCAR have similar precision with SSDT-Net, but supports both RGB and Hyperspectral/False-Color scenarios. In practical, it is suggested that authors share more real use cases and explain why these two algorithms can not fit for those cases and why more precision is necessary for the potential return value and corresponding costs.
2. Please share the detail ideas of why compare DP@20P with other algorithms. Which kind of scenarios will fit in the DP@20P, and which are not fit.
Author Response
Thank you very much for handling our manuscript remotesensing-1750257 and for giving us thoughtful and constructive comments and suggestions. Please see the attachment for responses to the comments.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have replied to my comments and revised the manuscript, so it may now be accepted after minor revisions (paper would still benefit for a more indepth comparison with other works in the sector). I would still recommend having a native speaker examine your English grammar and style or employing MDPI's services. There is no need to re-review.
Author Response
Thank you very much for handling our manuscript remotesensing-1750257 and for giving us thoughtful and constructive comments and suggestions. Please see the attachment for responses to the comments.
Author Response File: Author Response.pdf