Subjectively Measured Streetscape Perceptions to Inform Urban Design Strategies for Shanghai

Round 1

Reviewer 1 Report

The article discusses subjectively measured streetscape perceptions to inform urban design strategies for Shanghai. The topic is worthy of research; however, a major revision of the manuscript is needed. In the current state of the manuscript, it appears to be a draft, is not well detailed, and is not in the format requested by the journal. The procedure is not detailed. Some major comments/changes need to be addressed in order to be considered for publication.

General comments

What are the advantages and disadvantages of the methodology used in this study?

Authors could also emphasize particular strengths of the study for potential applications of their method in other regions, contexts and scales.

The answer to these questions should be reflected in the manuscript.

The data and methods section contains results, it is necessary to define a unique section for results.

Specific comments

Line 27: Not the correct way to cite. Revise the authors' guide and change throughout the manuscript.

Line 52: How much was the low-throughput?

Line 59: This is the first time it is mentioned in the ML introduction, write in full.

Line98-99: "performed well", this how much is it?, is it a percentage?, a value of an indicator?, how much?. "They worked well”, compared to what?".

Line 100: And what did you find?

Line 120: Change "Fig. 1a" to "Figure 1". Apply throughout the manuscript.

Line 124: What is POI?

Line 129-131: Which urban designers and planners? Cite.

Line 138-139: Use the same font as the rest of the manuscript.

Line 141: Improve the figure. Can't make it out, raise its resolution.

Line 144: Place a map of the study area with its basic elements, north arrow, scale bar, coordinate grid and a macro-location map.

Line 153-154: Provide the program link.

Line 159: Improve the figure. Can't make it out, raise its resolution.

Line 174: Improve the figure. Unable to distinguish, raise resolution.

Line 197: To ensure the reliability of the data shown in the study, it is necessary to improve the figure. The legend of the figures cannot be distinguished and the reader will not be able to understand the content of the figure. Raise its resolution.

Line 221-229: Results in Data and Methods?

Line 225: Change "R2" to "R²".

Line 261: Remove the title "Top 15 important features predicting perceived scores".

Line 297: Why these 5 cities? Justify.

Line 304: How much is this "smaller variance", where can it be looked up?

Line 386: This looks like a discussion and not the conclusions of what was obtained in this study.

Line 430: The references do not follow the authors' guide.

Author Response

Thank you for your careful consideration and detailed comments. We have largely revised our manuscript per your suggestions, as follows:

 

Specific comments

Line 27: Not the correct way to cite. Revise the authors' guide and change throughout the manuscript.

Thanks, we have changed the reference style to IEEE (e.g., [1], [1-3]), starting from line 35 throughout the paper.

Line 52: How much was the low-throughput?

We added details, e.g., “it took an hour to rate a single video clip”. See line 69.

Line 59: This is the first time it is mentioned in the ML introduction, write in full.

Thanks. See line 77. We have changed it. Although in the abstract we have mentioned ML in full.

Line98-99: "performed well", this how much is it?, is it a percentage?, a value of an indicator?, how much?. "They worked well”, compared to what?".

See line 121-123, e.g., “90% precision”, and they found “visual enclosure was significantly negatively associated with pedestrian count and the walk score”.

Line 100: And what did you find?

See line 123-128. Other researchers have classified various features from SVIs. However, the most important argument we addressed is that “for perceptual qualities that are not familiar to the average person, such as Imageability, a subjective framework exhibits better performance”.

Line 120: Change "Fig. 1a" to "Figure 1". Apply throughout the manuscript.

Thank you, we have changed them accordingly. For example, see line 145.

Line 124: What is POI?

Thanks, we added the description for POI. See line 149-151.

Line 129-131: Which urban designers and planners? Cite.

Thanks, we cited them in line 174, i.e., [2]. We also expanded the discussion in line 171-182.

Line 138-139: Use the same font as the rest of the manuscript.

Have changed it and also check all fonts throughout the paper, thank you.

Line 141: Improve the figure. Can't make it out, raise its resolution.

Thank you, we improved all figures. For example, Figure 1 is separated to two lines, see line 158.

Line 144: Place a map of the study area with its basic elements, north arrow, scale bar, coordinate grid and a macro-location map.

Thank you, have add the location map, north arrow, scale bar in Figure 2a. See line 189- 191.

Line 153-154: Provide the program link.

Thank you, we added the links for all API services, see line 167-169.

Line 159: Improve the figure. Can't make it out, raise its resolution.

Thank you, we enlarged it as well as all other figures throughout the paper. For Figure 2, see line 189.

Line 174: Improve the figure. Unable to distinguish, raise resolution.

Thank you, for Figure 3, see line 265.

Line 197: To ensure the reliability of the data shown in the study, it is necessary to improve the figure. The legend of the figures cannot be distinguished, and the reader will not be able to understand the content of the figure. Raise its resolution.

Thank you, we enlarged all figures. For Figure 4, see line 301.

Line 221-229: Results in Data and Methods?

Thank you, we have largely revised this section, now all results were moved to section 4.2. “ML prediction model performances” starting from line 547.

Line 225: Change "R2" to "R²".

Thank you, have changed it and all other places, see line 549, 576, and 705.

Line 261: Remove the title "Top 15 important features predicting perceived scores".

We removed it, see line 545 for Figure 5.

 

Line 297: Why these 5 cities? Justify.

Thank you, this should have been addressed. See section 3.4 “Global comparison with other cities” starting from line 411 to 431.

Line 304: How much is this "smaller variance", where can it be looked up?

Thank you, we added a Table 4 (line 657) to report the variance, standard deviation and mean values. See line 634-637.

Line 386: This looks like a discussion and not the conclusions of what was obtained in this study.

We moved the discussion to corresponding place in the discussion section. Regarding the rest (i.e., the forth conclusion) on the advantages of our method compared to other deep learning frameworks, we kept it in the conclusion section.

Line 430: The references do not follow the authors' guide.

Thank you, have changed them to IEEE style, see line 800.

Author Response File: Author Response.docx

Reviewer 2 Report

I am glad to have the opportunity to review this manuscript. The authors have developed an interesting solution to a significant problem in street view evaluation. The overall methodology of this study is inspiring. However, there are a few comments that needs the authors’ attention before the publication of this manuscript.

1. Figure 1 is somewhat blurred, the methodology of this study should be explained in detail in order to allow the potential readers to repeat your study and understand your contribution.
2. The four indexes should be further explained. How did the volunteers evaluate the four indexes in a subjective manner? For example, how did they rate for “human scale” in a given street scene?
3. Why do you perform the feature classification after the evaluation? I think it would be much better if you perform the feature classification first, and ask to volunteers to not only compare street scenes, but also select the most attractive features from the scene.
4. Following the last question, in general, the results in Table 1 is not fully explained, the logic flow of how these features impact the ratings should be revealed – we cannot rely on the black box of machine learning to do such explanation.
5. All the figures in this manuscript should be improved for better reading experience.
6. The discussion section should be further expand, especially on the performance of the ML and drawback of this study.

Author Response

Thank you so much for the helpful comments. We have largely revised our paper accordingly. 

 

1. Figure 1 is somewhat blurred; the methodology of this study should be explained in detail to allow the potential readers to repeat your study and understand your contribution.

Thank you, we have revised and enlarged all figures, especially Figure 1. See line 158- 159.

 

2. The four indexes should be further explained. How did the volunteers evaluate the four indexes in a subjective manner? For example, how did they rate for “human scale” in a given street scene?

Thank you, we addressed it 3.2.2 Collecting public perceptions as training labels, in line 231- 303.

 

3. Why do you perform the feature classification after the evaluation? I think it would be much better if you perform the feature classification first and ask to volunteers to not only compare street scenes, but also select the most attractive features from the scene.

Thank you, what you point out could be a future study that we can address in a future study. Have incorporated it in section 5.2 Limitations section. See line 805- 809.

 

Meanwhile, we explained the rationality in 3.2.2 Collecting public perceptions as training labels section.

 

We kept the original method and procedure, setting the framework based on what Ewing & Handy (2009) & Naik et al., (2014) took: asking people to rate an image (Naik et al., 2014) or a video clip (Ewing & Handy, 2009) directly. It is because viewers rate an image based on the overall/ comprehensive image appearance. The feature segmentation/ classification is to construct explanatory variables that built on classical urban design theory (Ewing & Handy, 2009).

 

In general, we chose this because human perception is comprehensive and subtle, viewers perceive a scene through the overall sensory information obtained by looking at the whole picture. Therefore, instead of addressing a particular segment/ part/ feature that stand out, we care about people’s overall understanding more.

 

4. Following the last question, in general, the results in Table 1 is not fully explained, the logic flow of how these features impact the ratings should be revealed – we cannot rely on the black box of machine learning to do such explanation.

Thanks, we largely revised this part.

 

First, we explained the rationality of keeping features extracted to become explanatory variables in predicting perceptions in section 3.2.4 “Streetscape feature selection” in line 304- 317.

 

Second, we explained the selection of ML models and their suitability in a separated section 3.2.5. “Predicting subjective scores”, see line 318- 401.

Third, the result discussion including Table 1 were relocated to section 4.1 “Descriptive statistics of the classification and significant streetscape features” and expanded, see line 437- 546.

 

Hope it works.

 

5. All the figures in this manuscript should be improved for better reading experience.

Thanks, we have improved all figures throughout the paper.

 

6. The discussion section should be further expanded, especially on the performance of the ML and drawback of this study.

Really appreciate this suggestion. We added discussions on ML performance to a section 4.2 see line 547-581. The drawbacks were expanded incorporating your comments on the classification and survey design in line 756-774 in section 5.2. Limitations

 

Author Response File: Author Response.docx

Reviewer 3 Report

The paper deals with applying computer vision to street view imagery dataset to extract the view indices of various streetscape features. For that purpose crowdsourcing, computer vision, and machine learning were used to subjectively measure four important perceptions suggested by classical urban design theory. As can be seen in conclusion, the researchers got to the point. The research is well enough conducted from my point of view and can be published.

There are a lot of small errors through the paper  that should be corrected. For instance:

Line 23, comas instead of dots, capital letters only where needed.

Line 81 space

Line 138, 139 different letter hight

Figure 1 is not readable. Suggestion is to enlarge them and put them as two separate figures

On Figure 2, 3, 4, 7, 8, 9, 10, 11 text is again too small to read, all should be bigger, photos and text.

Line 425 space missing

Line 427 da-taset

Check all the paper with spelling and grammar.

Need graphical improvement.

Author Response

Thank you for the detailed comments and we have revised our manuscript accordingly. Hope it is in better shape now. Best regards

Q: Line 23, comas instead of dots, capital letters only where needed.

A: First of all, thank you so much for the careful reading.

Q: We have changed the keywords’ format accordingly. See line 31.

Q: Line 81 space

A: Thank you, we have added the space.

Q: Line 138, 139 different letter heights

A: Thank you, have formatted it. We also reformatted the whole paper using the IJGI template.

Q: Figure 1 is not readable. Suggestion is to enlarge them and put them as two separate figures

A: Thank you, we have enlarged it and separated them to two separate lines. See line 158.

Q: On Figure 2, 3, 4, 7, 8, 9, 10, 11 text is again too small to read, all should be bigger, photos and text.

A: Thank you, we have enlarged all figures.

Q: Line 425 space missing.

A: Thank you, we have added the space.

Q: Line 427 “da-taset”.

A: Thank you, we have changed it to “dataset”. See line 796.

Q: Check all the paper with spelling and grammar.

A: Thank you, we have worked on this throughout the paper.

Q: Need graphical improvement.

A: Thank you, we have polished all figures, especially have enlarged the texts in all figures.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

please see the attachment.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

We really appreciate your careful reviews. All comments are super helpful to improve both our content and the format. We largely and carefully revised the format especially the citation and reference per MDPI style. We also added details accordingly per your suggestion and question.

It has been a very helpful learning process for us during the first and second rounds of revision. Thank you!

Author Response File: Author Response.docx

Reviewer 2 Report

The authors have improved the manuscript. It should be accepted for publication. 

Author Response

Thank you so much, we appreciate the support and the feedback received during the first round of revision. It has been super helpful, learned a lot. We made several improvements regarding the details, the reference, and the citation format.

Author Response File: Author Response.docx