9. Images, Art & Video - [E3] 3. Naik, Nikhil, Scott Duke Kominers, Ramesh Raskar, Edward L. Glaeser, César A. Hidalgo. 2017.

[JunsolKim]

Post questions here for this week's exemplary readings: 3. Naik, Nikhil, Scott Duke Kominers, Ramesh Raskar, Edward L. Glaeser, César A. Hidalgo. 2017. “Computer vision uncovers predictors of physical urban change.” PNAS 114(29):7571–7576.

[facundosuenzo]

I liked the paper; it is innovative, concise, and straightforward. However, it introduces a computer vision method to test already established theories but does not propose too many variations or explanations. I'm particularly unsure about the validity of the "safety" construct as a proxy for neighborhood "improvement", considering that neighborhood improvement may contain multiple variables beyond the perception - binary in this case - of safety.

[YileC928]

I liked the paper; it is innovative, concise, and straightforward. However, it introduces a computer vision method to test already established theories but does not propose too many variations or explanations. I'm particularly unsure about the validity of the "safety" construct as a proxy for neighborhood "improvement", considering that neighborhood improvement may contain multiple variables beyond the perception - binary in this case - of safety.

I second Facundo's point. I feel that 'safety' only captures one aspect of 'urban improvement'. Since the authors are interested in improvements in general, they could have also considered other features such as convenience and modernization. Would it be better if we construct a comprehensive index for 'streetscore'?

[Hongkai040]

The world of this study is meaningful. However, I am wondering if it's appropriate to use census data from 2000 along with pictures from 2007 and 2014 for analysis? The time gap could make the results obtained less reliable. As for pictures, they're cropped from Google Street View which is originally intended for guidance. They're taken from roads! For those changes happened away from roads, I'm afraid that those photos are unable to capture. Maybe this is the best dataset authors can get, but maybe these factors could influence their results they get.

[ZacharyHinds]

This is a really interesting study, but I echo the concerns about the limited scope that safety has in capturing "urban improvement" especially because determining safety based on images seems likely to be influenced by our biases of what features mark an area as "unsafe" (which may just be the signs of poverty in general). To reiterate YileC928's point, if we believe that such an index could work better, how could we develop that index?

[melody1126]

The data samples used in this study contained Google Street View images of streets in Baltimore, Boston, Detroit, New York, and Washington, DC. However, these cities have very different urban landscapes and regional / metropolitan planning diagrams amongst themselves and also when compared to other US cities like Chicago, San Francisco, Miami, Portland, and so on. I am curious about the external validity of this study's results and how applicable these labelled images map onto the socioeconomic and demographic information of a specific urban area.