Critique: Google Map's Moat

This post is a critique of Justin O’Beirne’s blog post “Google Map’s Moat: How Far Ahead of Apple Maps is Google Maps?” found here. Justin O’Beirne wrote this sometime late last year (2017), and I chose to critique (and share) this article because of how interesting I think its finding, that Google is creating data out of data, is.

Summary of “Google Map’s Moat”

Note: all of the images and GIFs used in this section are taken directly from the article. The context of the article is that, as of late 2017, Google Maps was considered far superior as a service to Apple Maps. Justin compares both services’ maps for several areas, from large metropolitan areas like New York City to small towns with populations fewer than a thousand people, and from the screenshots, it is abundantly obvious that Google’s maps are far more advanced than Apple’s. Whereas Apple Maps usually has accurate outlines of roads—traditional road maps—Google Maps is incredibly detailed, accurately showing not only the roads but also architecturally-similar 3D models of buildings (including unique building features). Relative to Google Maps, Apple Maps looks strikingly sparse and empty.

Google Maps even models “structures” like garages, tool sheds, and park shelters.

Here’s an example of the unique building feature mentioned earlier. In the first case, the building feature is the strange roof of the author’s local post office, and in the second, it is the HVAC units on the building’s roofs.

While Google Maps has these incredibly detailed building models for small towns of even a hundred people, Apple Maps doesn’t even have rough building models for major metropolitan areas.

Google revealed how they were generating these models, as early as 2012: “These building footprints, complete with height detail, are algorithmically created by taking aerial imagery and using computer vision techniques to render the building shapes…For landmarks like Seattle’s Space Needle, computer vision techniques extract detailed 3D models.” The author provides this flowchart to explain how Google Maps uses its satellite imagery to create detailed maps.

(This is where it gets really interesting.)

People tend to visualize cities in subsections, e.g. business and financial districts, downtown, etc. During their research, a couple graduate students realized this when their subjects would divide San Francisco into “mini-cities,” visualized below.

Google Maps visualizes this through its highlighted “Areas of Interest,” displayed below.

The graduate students spent months creating these subsections of San Francisco, but Google was able to do it programmatically for thousands of cities, metropolitan and rural alike. The author then goes on to discover Google’s Areas of Interest coincide with its buildings, and that Google is creating them by matching its building models (from satellite view) and place data (from street view) together to algorithmically generate the areas of the city with highest “interest” density. In other words, Google is creating data out of data.

The rest of the article ties this back to how seemingly insurmountable this technological gap is between Google and Apple Maps, but that’s not relevant to what I want to talk about.

Critique

This isn’t really a critique of the article as much as it is just a statement of admiration. To create data out of data like the author explained in the article is honestly mind-blowing. It implies an incredible level of accuracy for the original datasets using very experimental fields of computer science like computer vision. For Google to achieve such accuracy is fascinating, and most of the reason I chose this article to critique is because, as I mentioned earlier, this is so interesting I wanted to share it on my website.

To the author’s credit, it’s a well-written article that gradually guides the reader through his discovery process. The images and GIFs I’ve included in the summary are only a subset of the article’s. In fact, there’s an explanatory piece of media, typically an image or GIF comparing Google/Apple Maps or a time lapse of how Google Maps has become more computationally advanced over the years, after each statement or step through the discovery process, and the article is generally very accessible; it minimizes the technicality of the technologies behind Google Maps and only explores the results and outputs of these technologies.