SV-FPN: Small Object Feature Enhancement and Variance-Guided RoI Fusion for Feature Pyramid Networks

Round 1

Reviewer 1 Report

paper needs major changes

 

a)     What is at Y-axis in Figure 5? authors must clearly put the title with unit at both Y and X-axes

 

b)    The quality of Figs 1 and 4 is poor so can be redrawn with high visibility and resolution

 

c)     The introduction section can be rewritten with clear contribution and significance.

 

d)    Authors must develop the framework/architecture of the proposed methods

 

e)     There is need of flowchart and pseudocode of the proposed techniques

 

f)     Proposed methods should be compared with the state-of-the-art existing techniques

 

g)    Research gaps, objectives of the proposed work should be clearly justified.

 

h)    To improve the Related Work and Introduction sections authors are highly recommended to consider these high quality research works <A Multi-sensor Data Fusion Enabled Ensemble Approach for Medical Data from Body Sensor Networks’>, <Dynamic Application Partitioning and Task Scheduling Secure Schemes for Bio-Sensors Healthcare Workload in Mobile Edge Cloud >

 

i)      English must be revised throughout the manuscript.

 

j)      Limitations and Highlights of the proposed methods must be addressed properly

 

 

k)    Results section is weak, so more results can be added

 

 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper proposes a small object feature enhancement and a variance-guided region of interest fusion for improved small object detection. The results support the claim of improvement in 3 different datasets measured by average precision for 3 different object sizes, especially for the small class. I believe it is a good contribution on small object detection.

 

Minor issues:

 

- Equations should be part of the text and not cited in its first appearance.

 

- Define variables of equation (2).

 

- There is no need to place multiplication operators in equations, and indeed an asterisk may be misunderstood as convolution instead of multiplication.

 

- Table 1 last column: either first row is \leq instead of \geq and last row > instead of <, or the values 14 and 28 are swapped between first and last rows.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper is required extensive review on all aspects. For instance there are many grammatical errors. Go through again all words line by line.

The literature review sections is no t comprehensive. 

The methodology also not clear convincing.

Experimentation and Results are not comprehensive.

Validation is not adequate.

Based on above matter, I reject the paper. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Authors have improved paper at some extent, but still there are major changes to be fixed

Reviews to Authors

1. Resutls section is still weak, so more resutls can be added 

2. Authors are highly suggested to give more details on the propsoed method with clear flowchart, working principle, and pseudocode for better and clear understanding

3. Number such as (1), (2).....must be repalced with bullets and First, ......, Second, ......, in the 'our contribution' to avoid the confusion with numbering of main sections

4. Related work section must be extened by considering high quality works <A Multi-sensor Data Fusion Enabled Ensemble Approach for Medical Data from Body Sensor Networks’>, for detailed and better insight to the readers

5. More information and details are required in the captions of all the figures and table titles

6. Limitations , research gaps of the existing works must be clearly addessed for understanding the need and importance of the propsoed research 

Paper needs major changes 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Slightly improved than earlier version. 

Please add algorithm and if possible link for the coding that can be used for the implementation. 

Go through English one more time. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Reviews for Authors

1. Title is still longer so must be rewritten in a concise and clear way

2. What is relation between Figure 7 and Figure 8? there seems no any connection and flow between results.

3. Figure 7 seems strange with only Ps, it shall be good to establish the connection between all the results for better and clear understanding 

4. Related work section is weak so authors are highly suggested to add these highl quality works <A Multi-sensor Data Fusion Enabled Ensemble Approach for Medical Data from Body Sensor Networks’>, because for the clear and indepth understanding and connection with existing works is vital

5. Limitations of the propsoed method must be rewritten with better future recommendations in the conclusion section