I hope this and Scape are tools for a successor to PhotoSynth. It's really a shame that that dream from 2007-2010 era never materialized. For facebook to be able to turn all its photos of public places into geometry and texture sets, and procedurally create 3D representations of the world at ground level, without street view cars, would be a next big moment in consumer mapping, and a win for Occulus. Textures in particular would look much better being extrapolated and generated vs being overlayed from the photos themselves, and lighting could be something reverse calculated from photos at different times of day. Imagine being able to turn a photo from day to night, based on what you know about said locations actual lighting, and not a generic filter. Those 3D maps would also be handy for AR, being able to sync the geometry of the world up to what facebook expects it to be.
This is the future we were/are looking to build with https://ayvri.com, but you have to start with a compelling use case first. It's great to be able to create a this incredibly cool tech, but what will people do with it? If you look at Google Earth, it's amazing, it's cool, but so many years in, there still is barely a use-case for the amount of $$ pumped into it.
We're fortunate to be profitable, and have users that love what we do today, but it's a long long road to get to where we want to be.
The vision as I was explaining it in my last funding round was to imagine a world where every photo or video you took had infinite zoom and rotation of every photo or video you took.
Mostly that vision resulted in a ton of push-back on how it will never be possible, which is BS based on the progress of computer vision and AI in computer graphics, or "so what".
I'm thinking the value of having Yelp overlayed on VR/AR alone.
Youre sitting at home, and you do a walkthrough of your vacation before you go, get familiar with the location, learn your routes. Then you get there, and you have your map/dots/itinerary overlayed.
Being able to walk around an area, to find things that look interesting to you, vs searching yelp/tripadvisor.
I'm on board with your vision. Realtime infinite zoom and synthetic upscaling are around the corner. Sure you wont be able to add actual detail of specific things that isnt captured anywhere, but it will be an acceptable substitute. Being able to reposition a camera in a photo, after it is taken.
On the "actual detail of specific things" comment, I think generative AI will get a long way there to be convincing. But almost more importantly there are two non-obvious things that people don't realize when thinking of why this won't work.
1) the camera is focused on the subject mater, and we don't offer free roam. The subject is always the focus and we more around that subject. Because the subject matter is focused, the real-world camera never captured that information anyway. The person taking the photo or the person viewing it, doesn't know what is there and what isn't, so to convince them of what they are seeing, as long as the subject matter is re-created, should be fine for 90% of the content. We're not re-creating crime scenes, we're giving you a new way to experience your memories.
2) we mis-remember so many things anyway, that these details that (as mentioned above are likely not important) will the viewer remember that the person walking past behind the person taking the photo had a blue shirt on or yellow. Will it matter? We already have a flexible relationship with media, as so much is photoshopped to get the image the person wanted to take vs what they actually took. I suspect we will go deeper down this rabbit-hole and be more and more and expect to be thinking to ourselves "it's a nice photo, probably isn't like that IRL".
yeah, most people I ask about Google Earth say "I looked at my house", or "I'm going on vacation and checked out the location".
None of these are "killer apps", or "daily use" features. After you've got selected your college, is there anything that you will use google earth for afterward?
It's an AMAZING product, I'm not taking away from the product, but the use doesn't justify the effort.
I'm currently using Street View quite a lot to check the surroundings of houses I'm interested in. I think this is something a potential buyer could pay a little for but real estate agents usually do their best to hide the actual location of the house (maybe they fear to be bypassed?)
Other use cases:
I check roads, intersections and landmarks when I'm planning new bicycle rides. Satellite view is also very useful. I do most of the planning on the OSMAnd app and I check some parts on Google.
I check the locations I must go to of they are new to me. This is to know where to go and not lose my way on the last few meters (street numbers sometimes are unhelpful.)
I use satellite/aerial for cycling, as it's updated quite frequently and it's reasonably easy to tell apart the bad path (sand/big traffic). The Street View in the less crowded places seems to get one update in 8-10 years.
Loosely related, but I was thinking about how Facebook has a treasure trove of information with regard to backcountry data such as waterfalls and sensitive rock art sites.
Many people in my area post photos of these sites to Instagram without a location tagged, but they likely upload with the actual location data still in the image metadata. Because of this you could put together some pretty comprehensive glimpses at these hotspots.
Mapillary is deeply connected to OpenStreetMap. Google steers clear of OSM-related things in order to protect its own maps offering. There is no way Google would have bought Mapillary out except maybe as an acquihire, and even then, did Mapillary employees have any unique skills that Google hadn’t already had in abundance after a decade of Street View development?
19 comments
[ 5.4 ms ] story [ 54.3 ms ] threadThese original two ted talks were spectacular.
https://www.ted.com/speakers/blaise_aguera_y_arcas
https://www.wired.com/2010/11/mining-flickr-to-build-3d-mode...
We're fortunate to be profitable, and have users that love what we do today, but it's a long long road to get to where we want to be.
The vision as I was explaining it in my last funding round was to imagine a world where every photo or video you took had infinite zoom and rotation of every photo or video you took.
Mostly that vision resulted in a ton of push-back on how it will never be possible, which is BS based on the progress of computer vision and AI in computer graphics, or "so what".
Youre sitting at home, and you do a walkthrough of your vacation before you go, get familiar with the location, learn your routes. Then you get there, and you have your map/dots/itinerary overlayed.
Being able to walk around an area, to find things that look interesting to you, vs searching yelp/tripadvisor.
I'm on board with your vision. Realtime infinite zoom and synthetic upscaling are around the corner. Sure you wont be able to add actual detail of specific things that isnt captured anywhere, but it will be an acceptable substitute. Being able to reposition a camera in a photo, after it is taken.
1) the camera is focused on the subject mater, and we don't offer free roam. The subject is always the focus and we more around that subject. Because the subject matter is focused, the real-world camera never captured that information anyway. The person taking the photo or the person viewing it, doesn't know what is there and what isn't, so to convince them of what they are seeing, as long as the subject matter is re-created, should be fine for 90% of the content. We're not re-creating crime scenes, we're giving you a new way to experience your memories.
2) we mis-remember so many things anyway, that these details that (as mentioned above are likely not important) will the viewer remember that the person walking past behind the person taking the photo had a blue shirt on or yellow. Will it matter? We already have a flexible relationship with media, as so much is photoshopped to get the image the person wanted to take vs what they actually took. I suspect we will go deeper down this rabbit-hole and be more and more and expect to be thinking to ourselves "it's a nice photo, probably isn't like that IRL".
probably unimportant, I use it for college tours and renting houses
It's an AMAZING product, I'm not taking away from the product, but the use doesn't justify the effort.
Other use cases:
I check roads, intersections and landmarks when I'm planning new bicycle rides. Satellite view is also very useful. I do most of the planning on the OSMAnd app and I check some parts on Google.
I check the locations I must go to of they are new to me. This is to know where to go and not lose my way on the last few meters (street numbers sometimes are unhelpful.)
Many people in my area post photos of these sites to Instagram without a location tagged, but they likely upload with the actual location data still in the image metadata. Because of this you could put together some pretty comprehensive glimpses at these hotspots.