Next Article in Journal
Vec2Dynamics: A Temporal Word Embedding Approach to Exploring the Dynamics of Scientific Keywords—Machine Learning as a Case Study
Previous Article in Journal
The Next-Generation NIDS Platform: Cloud-Based Snort NIDS Using Containers and Big Data
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
BDCC
Volume 6
Issue 1
10.3390/bdcc6010020
Review Report
BDCC-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
Article
Peer-Review Record

Person Re-Identification via Pyramid Multipart Features and Multi-Attention Framework

Big Data Cogn. Comput. 2022, 6(1), 20; https://doi.org/10.3390/bdcc6010020
by Randa Mohamed Bayoumi 1,*, Elsayed E. Hemayed 2,3, Mohammad Ehab Ragab 1 and Magda B. Fayek 2
Reviewer 1: Anonymous
Reviewer 2:
Radosław Roszczyk
Big Data Cogn. Comput. 2022, 6(1), 20; https://doi.org/10.3390/bdcc6010020
Submission received: 31 December 2021 / Revised: 27 January 2022 / Accepted: 29 January 2022 / Published: 9 February 2022
(This article belongs to the Topic Machine and Deep Learning)

Round 1

Reviewer 1 Report

The paper proposes a deep learning framework for person re-identification that relies on multi-part features and multiple attention to aggregate features captured in multi-levels. The basis of the framework is the self-attention mechanism that tries to strengthen the most representative features. The framework achieved 98.9% mAP on PRID2011 dataset and 92.8% Top1 Acc on iLids-vid dataset.  

 

The paper overall shows a very interesting approach that was tested in 4 different datasets and the results were compared against several other proposals.

 

My overall suggestions are:

  • The text should be reviewed and the authors should keep a uniformity of what should be in capital letters and what should not.
  • In Table 2, why there is "BICnet" if there is no value associated with it? 
  • The model proposed is very complex, would be great to see a table with the training time for each dataset. Moreover, if possible, try to compare this time against the other models evaluated. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is interesting and raises the interesting issue of re-identification of persons. It can be classified as a paper presenting the application of neural networks in practical solutions. The authors of the paper presented many different alternative solutions to the problem and showed that their solution is slightly better than the others.

In the conclusions, the purpose of the developed algorithm and the possibilities for its use should be clearly highlighted. It is also worth emphasising what the potential advantage of the solution over others will be. The percentages results do not fully represent the possible benefits of the approach.

Amendments to be made:

  • It would be useful to include and comment on an instance with several people in one image.
  • Language corrections are required

  •  

    In Figure 3 and 4, the description of the input features (N xt) x h x w x c is illegible.

  •  

    Figure 5 - the input signal to the aggregation layer is unclear.
  •  

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Back to TopTop