Utilizing Mapillary (in addition to or instead of) GSV

[jonfroehlich]

We've long wanted to be able to support multiple streetscape datasets (e.g., from Mapillary, Apple Maps, Bing Maps).

I spent a bit of time today (while multitasking on Zoom calls) exploring Mapillary.

Mapillary has many advantages—it relies on crowdsourced streetscape data; however, this data is far sparser, more diverse in quality, and not full 360.

Here's the current coverage map of Mapillary:

Here's the coverage map of Seattle. Looks good on a quick scan but you quickly see many streets without coverage (e.g., in Wallingford, Queen Anne, Ballard, Ravenna, etc.):

It's worse in less populated areas. Here's Bremerton, for example:

Here's Poulsbo:

Here's a comparison with GSV:

People contribute a variety of data from dashcams, mobile phone images, etc. Here's an example from a dashcam in Seattle but doesn't let you pan the image (so it's a fixed view pointing straight ahead)

So, my sense is that Mapillary is still not a good enough dataset for us.

On the other hand, it could be a possibility if people are uploading their own 360 images from on-the-ground audits.

[jonfroehlich]

Just quickly investigated Apple Look Around.

• Here's a Reddit thread about scraping the data. Here's some Python code on GitHub.
• Here's an interesting analysis of Apple's drivethroughs: https://www.justinobeirne.com/look-around-rollout. Some cool tidbits in here such as Apple's Look Around covering every metropolitan area in Canada (35 in total) + thousands of rural towns and villages between them. Look Around is also now available in all US counties