Sean Hardesty Lewis's VLM-on-Manhattan-street-view project; legal/TOS questions; alternative providers (CycloMedia, Mapillary, KartaView). Image data is a potential market expansion lever beyond NYC.
Marc (CRO) — 2026-01-22 21:22
This is cool. Not our immediate priority... but as we have discussed... images are essential to our work outside of NYC.... but based on this site.... still also relevant to nyc My gut feeling.... from a sales/buzz standpoint.... blending this kind of data with BKB would get us more widely noticed (also of course functional value)
"...ran a Vision Language Model across hundreds of thousands of Manhattan street-view images and asked a simple question:
"Describe what you see."
The result is the first open‑vocabulary semantic atlas of NYC.
Basically: A city you can Ctrl+F. "
https://www.linkedin.com/feed/update/urn:li:activity:7419378858853908480/
My favorite searches
"open windows in summer" --> most of manhattan lights up with exception of midtown and downtown blocks dominated by office buildings with inoperable windows
"open windows in winter" --> much more concentrated parts of Manhattan with higher concentrations of affordable housing.
But the "scaffolding" one is also relevant to our work
Marc (CRO) — 2026-02-23 16:57
@François (HoPlatform) so you were right... I cold-pinged the guy who built this... and he was kind enough to give me this background. Upshot: he thinks he can do whatever the fuck he wants for research (vs product) applications.
Me: I thought that doing this processing on google images was against their terms of service... am i missing something?
Sean Hardesty Lewis:
Good question! technically yes, it is a derived visualization using Google imagery through official APIs. However, will they do anything about it like send me a seize and desist? Probably not as long as I don't try to make it a product or compete with their business. Research has a kind of safety net that giant companies will sometimes be angry (since research labs can often form startups ex. Michael Bernstein's Simile or Feifei Li's World Labs) but often will be fine with accreditation since it just adds to their value (i.e. these researchers used Google data, so we should too)
Other projects similar to mine that use Google data and haven't been TOS'd yet are All Text in NYC by Yufeng Zhao (alltext.nyc) and my identical twin brother's Voxel Earth (voxelearth.org). We can also see quite a few papers on arxiv and in conferences like NeurIPS, AAAI, etc. that attribute Google and use their data for research and are often not funded or associated with them. Will Google TOS every single research project? Probably not.
Great question though! Appreciate you connecting with me
voxelearth.org
also I'm releasing all of NYC in the next few days (fixing up some UI bugs now) so if they want to TOS me they will likely do it when they see the entire city haha
Marc (CRO) — 2026-02-23 17:07
I do think images are critical. I do know we can't do it all at once. When we are ready to do images, there are paid sites that could be evaluated. But maybe we will be able to get going on this for free with NYSERDA data.
CycloMedia: A Dutch company that systematically captures 360° street imagery with high-resolution cameras mounted on vehicles.
Mapillary / KartaView (Crowdsourced but licensable data). Mapillary is a huge crowdsourced street imagery database now owned by Meta; you can license and use its data under CC BY-SA terms.
KartaView (formerly OpenStreetCam) is another open imagery source with similar use cases.
These aren't Google Street View — but they are "street-level imagery that you can build apps and analytics on" and are legally usable under their licenses.
François (HoPlatform) — 2026-02-23 17:29
Thanks for the follow-up Marc, that makes sense that they didn't get yelled at yet since they're not making money off of this.
Strategy Notebook
Sean Hardesty Lewis's VLM-on-Manhattan-street-view project; legal/TOS questions; alternative providers (CycloMedia, Mapillary, KartaView). Image data is a potential market expansion lever beyond NYC.
Images — Jan 22 – Feb 23, 2026 #
Source: #strategy-huddle
Marc (CRO) — 2026-01-22 21:22 This is cool. Not our immediate priority... but as we have discussed... images are essential to our work outside of NYC.... but based on this site.... still also relevant to nyc My gut feeling.... from a sales/buzz standpoint.... blending this kind of data with BKB would get us more widely noticed (also of course functional value)
"...ran a Vision Language Model across hundreds of thousands of Manhattan street-view images and asked a simple question: "Describe what you see." The result is the first open‑vocabulary semantic atlas of NYC. Basically: A city you can Ctrl+F. " https://www.linkedin.com/feed/update/urn:li:activity:7419378858853908480/
My favorite searches "open windows in summer" --> most of manhattan lights up with exception of midtown and downtown blocks dominated by office buildings with inoperable windows "open windows in winter" --> much more concentrated parts of Manhattan with higher concentrations of affordable housing. But the "scaffolding" one is also relevant to our work
Marc (CRO) — 2026-02-23 16:57 @François (HoPlatform) so you were right... I cold-pinged the guy who built this... and he was kind enough to give me this background. Upshot: he thinks he can do whatever the fuck he wants for research (vs product) applications.
Me: I thought that doing this processing on google images was against their terms of service... am i missing something?
Sean Hardesty Lewis: Good question! technically yes, it is a derived visualization using Google imagery through official APIs. However, will they do anything about it like send me a seize and desist? Probably not as long as I don't try to make it a product or compete with their business. Research has a kind of safety net that giant companies will sometimes be angry (since research labs can often form startups ex. Michael Bernstein's Simile or Feifei Li's World Labs) but often will be fine with accreditation since it just adds to their value (i.e. these researchers used Google data, so we should too)
Other projects similar to mine that use Google data and haven't been TOS'd yet are All Text in NYC by Yufeng Zhao (alltext.nyc) and my identical twin brother's Voxel Earth (voxelearth.org). We can also see quite a few papers on arxiv and in conferences like NeurIPS, AAAI, etc. that attribute Google and use their data for research and are often not funded or associated with them. Will Google TOS every single research project? Probably not.
Great question though! Appreciate you connecting with me
voxelearth.org
also I'm releasing all of NYC in the next few days (fixing up some UI bugs now) so if they want to TOS me they will likely do it when they see the entire city haha
Marc (CRO) — 2026-02-23 17:07 I do think images are critical. I do know we can't do it all at once. When we are ready to do images, there are paid sites that could be evaluated. But maybe we will be able to get going on this for free with NYSERDA data.
CycloMedia: A Dutch company that systematically captures 360° street imagery with high-resolution cameras mounted on vehicles.
Mapillary / KartaView (Crowdsourced but licensable data). Mapillary is a huge crowdsourced street imagery database now owned by Meta; you can license and use its data under CC BY-SA terms.
KartaView (formerly OpenStreetCam) is another open imagery source with similar use cases.
These aren't Google Street View — but they are "street-level imagery that you can build apps and analytics on" and are legally usable under their licenses.
François (HoPlatform) — 2026-02-23 17:29 Thanks for the follow-up Marc, that makes sense that they didn't get yelled at yet since they're not making money off of this.