Enhancement: SiamFC Tracker Algorithm Performance Based on Convolutional Hyperparameters Optimization and Low Pass Filter
Round 1
Reviewer 1 Report
The paper is well written but the reviewer has some comments.
1- In the related work section, why the authors have written some equations in the text lines as in line 144, 145 and 151. It should be in separate lines.
2- Some typos are found such as table should be Table in text and figure should be Figure please correct this in the whole paper.
3- Please Explain Figure 2 and Figure 3 in more details it is not clear for the reader.
4- What is HE in line 216 means please explain.
5- Please adapt the beginning phrase spacing in line 263.
6- At the end of line 292 the authors have written In order words it should be in other words.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This research proposes an algorithm / technique through which they tried to optimise and improve the performance of CNN. They did so by doing better initialization and activation function. They introduced the improved activation function named balanced ReLU based on which the output of CNN is improved. However there are few observations based on which they need to improve the article before further consideration. Observations are as follows:
- The dataset used, VOT2016, seems quite old, I would like authors to test there algorithm on some more recent dataset/s.
- Similarly the result comparison is only with a base paper [36], its good to see that the proposed results are improved with respect to [36]. But in order to establish the robustness and superiority of the proposed methods, authors must compare their work with atleast 1-2 more works other than [36].
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The suggestion is as follows
1.Highlight improvements in proposed work
2.Elaborate description of implementationim
3.Authors may focus on their highlights concerned with implementation
4.Detailed description for fig 2& 3 is required
5. Analysis to be done compared with existing algorithms
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
The paper focuses on improving FCSNN performances by optimizing hyperparameters and introducing low-pass filters. Some interesting elements are provided regarding the employment of Leaky ReLU and Gaussian filters.
Nevertheless, I believe the authors should push more on the novelty they introduce the existing architectures. For instance, the content overfocuses on some existing methods and highlights some implementation details. I think the paper should undergo sections' refurbishment.
My thought on contribution is that the current version does not provide enough novelty to the existing methods. That is due mainly to how the new elements are introduced in the paper.
If the main aim is to work out some improvements by adding or changing activation functions and using some low-pass filters, the paper should be focused on those elements.
I provide you with major and minor issues down below.
Major issues:
Section 2. The related works section overfocuses on two references: Bertinetto et al. and SiamFC. Instead, the authors should work on the content tackling more state-of-the-art techniques (for instance, FCNNs).
Section 3. I would invite the authors to focus on what is new in their paper concerning implementing the method in reference [36]. Is it only implementation work?
Could you please mention the differences between the approach in the following article and the Siamese Network? In the article, the employment of a Siamese CNN pulse some pre-processing functions based on Histogram analysis is carried out to have the training specialized on a specific application domain.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7528986/
Minor issues:
"The main contributions of our paper are the following:" (Line 98)
Could you rephrase the statement in line 112?
The following one should be more suitable:
For the last few years, target tracking has been one of the crucial areas in machine vision to 113 accomplish impressive progress.
I believe the statement in line 115 should be rephrased. Below is an alternative:
As an illustration, Hossein Kashiani et al. [26] introduce a 115 new approach that outperforms avant-garde trackers in the context of performance
Line 140: Did you mean "on which our work relies"?
Line 142: I find the statement in line 142 a bit unclear. What does "commuting beside translation" mean?
Line 148: "The analogy with" rather than "the analogy at"
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Accepted!
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
I want to express my appreciation for the authors' efforts to improve the paper's strength and content quality.
I believe that the paper in its current form is way better. It is more scientifically sound and easier to read. Well done on you.
I have only some minor comments pertaining to some latest changes I would suggest accomplishing.
Minor issues:
1. I think the introduction section lacks introductory elements on Siamese Networks, which are broadly employed in the computer vision community. To have readers better understand the main contributions of your paper from an architectural viewpoint, I believe you should tweak the corresponding part of the section.
For instance, I think the article below introduces a siamese CNN architecture aiming to test different existing CNNs such as AlexNet and PyramidNet to achieve various performances. I believe this should be part of the literature review about Siamese Networks.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7528986/
2. Regarding the state-of-the-art methods in object tracking, I would invite the authors to lay out some considerations on the trendy method, DeepSORT. It was widely employed for object tracking and paired with highly effective object detection models such as YOLO-v3, YOLO-v4 and YOLO-v5. The latter nowadays represents a pillar method for object tracking.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
