Improved Video Anomaly Detection with Dual Generators and Channel Attention

Round 1

Reviewer 1 Report

Interesting topic. Personally I prefer more examples from security/safety applications like drop object, taken object.... Presented situations are "very clean".

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Paper presents a study on video anomaly detection with dual generators.
#The description of the abstract is not sufficiently explanatory. From the abstract is not clear
what kind of new or novel research explored.
#There are lots of techniques available for video anomaly detection how dual generator superimposed on those?
#Authors have used publicly available datasets, CUHK Avenue and UCSD Ped1&Ped2. What if the database is biased as per the algorithm? Have you test this on some another dataset?
#How you are categorizing normal and abnormal events?
# What is the motivation of this paper?
# What is the contribution and novelty of this paper?
#What is the advantage of this paper?
#Which evaluation metrics did you used for comparison?
#For the network to converge better during training, you have used a constraint function to constrain the network. How you decide this constrain function will perform appropriately?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

- line 106: this method -> the proposed method

- line 114: Dual ~ -> dual ~

- It is not recommended to describe Figure 7 first in Section 4.

- Experimental results are not enough. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The manuscript is interesting and seems to have a contribution. However, there is a lack of detail in several sections of the manuscript, the English should also be improved. A PDF document with comments and recommendations is attached. The list of comments by section is below.

Introdcution:

The problem is statated correctly, however, its importance and applications are not clearly stated.

The caption of figure 1 should be self-contained.

In line 57, explain the term dual-generator. The usage of both "generator" and "generative" to name the full concept seems incorrect. I have found the term "dual generative adversarial networks" in the litarature, is this the actual term or is it a new approach?

Methods:

In line 57, explain the Dual-generator generative adversarial networks (DGGAN).

In figure 4, important terms such as "Conv 3x3", "BatchNorm3D" and "ReLU" are not explained in the caption.

In line 185, the term "second-order channel attention (SOCA)" is explained after its introduction on Figure 3, it should be introduced before.

In all the equations use homogeneous notation, and add references to all equations in the same format.

Experiments and results:

Explain the table 1 to te reader.

The term AUC is never explained.

In line 224, explain what AUC measures and how.

In line 242, "our method has a certain accuracy improvement in detecting anomalies" how much better compared to other methods?

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The figure 5 advised to include the explanation in the main text.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

The manuscript has been improved, the contribution is clear and all sections are now more detailed and consistent. A minor revision of the English language and style is suggested.

Author Response

Please see the attachment.

Author Response File: Author Response.docx