Open dobkeratops opened 5 years ago
Thanks a lot for sharing! I have to admit, I haven't heard of the company before, although they seem to have a pretty big dataset already. As far as I've seen from screenshots and videos, they have a pretty interesting user interface for annotating/labeling. I guess it's an advantage, that they focus on street images. That way they can design an UI that fits the use case perfectly.
While they have some really cool ideas (the map + the possibility to virtually "drive" along a specific highway section is totally awesome), there are also a few things that are a bit off putting to me:
all the images are CC-By-SA licensed : Not sure if this is a problem at all...the whole technology is probably still too new to tell...but I guess some intersting questions could arise in the future. ("If a neural net is trained on CC-By-SA licensed images, do I need to attribute every photographer?")
no public backups: Accoring to wikipedia, mapillary seems to be a full blown company that also cooperates with other big companies (like Amazon) to improve their services. While this alone is already off putting for me (I am personally not a big fan of those big data collecting companies, that only act in their own interest), I find it also a bit risky to contribute to their service. What happens when they can't sustain their business anymore? Will the service be shut down? What happens to my images then?
not fully open source: That's only a minor detail, but as far as I've seen, mapillary is not fully open source. They have quite a lot of open source libraries, but the core seems to be closed source (at least, I haven't found it).
But that's just my personal opinion. ;) I can totally understand when someone contributes to mapillary..they have a really nice UI and their maps feature kicks ass.
https://www.technologyreview.com/s/612825/open-source-maps-should-help-driverless-cars-navigate-our-cities-more-safely/?utm_medium=tr_social&utm_source=twitter&utm_campaign=site_visitor.unpaid.engagement
crowdsourced maps , interesting idea - I haven't read the article yet but it would be interesting to compare goals . The goal appears to be to use recognition of roadsigns (and landmarks?) to boost the accuracy of map position estimate from vision. I wonder if there are any inspirations here