RIIAnet: A Real-Time Segmentation Network Integrated with Multi-Type Features of Different Depths for Pavement Cracks

Round 1

Reviewer 1 Report

Hello Authors,

I have gone through the paper titled "RIIAnet: a real-time segmentation network integrated with multi-type features of different depts for pavement cracks". In this paper you have proposed a new network for pavement crack detection and the focus is more on the speed at which the network perform for real time image processing. 

Paper is interesting but I do have some minor comments to improve the paper quality:

- The use of work early diseases of road at several places does not jive well with the manuscript.  The term 'diseases' is typically used in medical terminology and use of the word here is not appropriate in my opinion. 

- Biggest criticism of your work is the sample size. You have used only 727 images and its not clear to what extent Crack 500 data set is used in the study. Have you tried to increase the number of images using image augmentation ? What is the statistics of images with different pavement types i.e. how many images are for asphalt how many for concrete and how many for tar ?

-  In section 3 of the data set different image numbers are mentioned in line 363 and line 765 and its not clarified if these images are form your image collection or from crack 500 data set ?

- Sample Images shown in Figure 7 have a good heterogeneity  but the images used for comparing different network shows in Figure 14 does not have similar heterogeneity. 

- The main advantage of the network seems to be the speed, which is a good point but it is also important to demonstrate that the speed of the network in classification task hold for high image heterogeneity.  I will suggest authors to elaborate on it in the revision either with help of an example or discussion. 

- Image heterogeneity due to impacts like shadows and challenges associated with it has been detailed in following publications, which should referenced in the paper:

https://www.mdpi.com/2076-3417/11/23/11396

 https://www.mdpi.com/1424-8220/22/10/3662

 

With these minor changes. I will be happy to recommend paper for publication.

Thanks

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this manuscript, the authors proposed an end-to-end real-time pavement crack segmentation network (RIIAnet) to inspect the pavement surface crack. The asymmetric convolution enhancement module (ACE) and the residual expanded involution module (REI) were used in the proposed network. Following comments are provided as a reference for the authors.

1. There are a lot of grammar and format mistakes in this draft. Please fix them

2. The Abstract is redundant, and it is difficult to find the authors’ contributions. What kind of issues do the authors expect to solve?

3. It is useless to describe the total highway mileage of U.S. and China. What’s the background? What are the current limitations? What are the motivations of this research?

4. Page 2. There is way more than three conventional methods to inspect the pavement crack. You can briefly introduce three types of conventional methods.

5. Page 3. A short literature review should focus on the previous contributions and limitations and then introduce the proposed method. It is NOT a retelling of previous articles. 

6. Page 4. “analyze” and “make suggestions” cannot be the contributions. It is a paper rather than a report. 

7. Figure 1. Please explain the design of different channel numbers. 

8. Figure 2. If the authors didn’t plot Figure 2, please cite the reference. There are different font types between Figure 1 and 2. Please use the same font size in the plots.

9. Figure 5. Why there are 1 by 3 convolutions? I know there are 1 by 1, 3 by 3, and 5 by 5 convolutions. 

10. Page 11. Why not use SGD as the optimizer? The batch size of 16 seems very high.

11. Figure 9. The feature maps are not convincing on feature extraction.

12. Table 1. Please unify the unit.

13. Table 4. It seems BiseNet has a better inference speed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

My comments have been addressed. It can be accepted after minor English editing.