This is really cool - now imagine what Google can do with project ATAP! [1]
If anyone knows anyone remotely involved with ATAP, can you please get them to get in contact with me. Google have inadvertently created a solution to the following scenario:
"Yes, place XYZ is completely wheelchair accessible"
"Oh, we forgot about ABC (step, bathroom, camber, terrain)"
Scan an area with ATAP and send it to me - I'll be able to tell whether it's really accessible or not.
Is added depth information really a game-changer? People have already used 2D, photosphere-style images to do what you're suggesting for streets and places visible on StreetView. [1] I understand that more information is going to be helpful, but are there many scenarios that significantly benefit from a depth map to judge accessibility?
>Is added depth information really a game-changer?
it is if you want to automate it instead of relying on meatbags.
Your link, just like many other research projects (LA pool database for example) rely on wetware looking at pictures. Depth maps let computers do all the work. You get (well, will get with tango) Quake style maps from simple video clips.
Looked at the code and got blown away - google is embedding depth map inside original images.
Google did something very smart here. They realized project tango is a long way from shipping, so they released something not perfect, clunky, manually operated, but working _right now_. This way they can claim to be first when new iphone with integrated primesense 3d camera hits the market :).
Can't remember where I read that images to 3D is hard, but this application really gives a good visualization of that problem. It's not the obvious parts that are difficult, but the relatively flat surfaces that are a little distant that cause the big issues.
This is amazing! I wonder how well distance accuracy would compare to something like the structure.io sensor attachment (same as in project ATAP). I wonder if there are other applications to this more coarse/less accurate depth information.
It seems like cool technology - it'd be a shame for it to fall by the wayside because people think software can achieve the same (although if I'm wrong, and it can, please correct me).
I have a Lytro and Nexus 5, and the biggest immediate difference in the user experience is that the Lytro captures all of the data in a single exposure. The Google Camera lens blur feature gets its depth data from having the photo-taker take a photo, then move the camera up (several inches to a couple feet, it seems, depending on subject distance). So if the subject is moving or you have unsteady hands, the capture will be cancelled.
So the technique to take the photo is very similar to the Seene app on iPhone that takes 3D photos. So the same technology could be potentially used for lens blur?
Seene website: http://seene.co
I think mainly because there isn't much of a usecase for most people.
We use Structure from Motion techniques a lot here where I work (Research Institute) to construct 3D models of buildings and land from UAV imagery.
The current methodologies are great for this use, but they lack the accuracy you'd want to model an object on your desk for 3D printing. That's not to say you can't do it - there's just often a lot of work to be done on the mesh afterwards to make it work.
Autodesk make a pretty good app called 123D catch [0], which uploads your images to their servers for processing.
Alternatively VisualSFM[1] is what I mostly use, I get great results but I don't know how good it is with small things.
I've been doing this stuff for a while. I can make 3d models from photos using VisualSFM and CMPMVS. It produces good results. I guess it hasn't taken off because for most people this is not cool. What does the average person do with this tech? Making mini statues of yourself is a nice application but most people find it a little creepy - not cool. I suppose it's like when photos started being a thing - it took a while for people to accept it. It took money and power to force it into peoples consciousness so that they could appreciate it.
15 comments
[ 2.8 ms ] story [ 44.5 ms ] threadIf anyone knows anyone remotely involved with ATAP, can you please get them to get in contact with me. Google have inadvertently created a solution to the following scenario:
"Yes, place XYZ is completely wheelchair accessible" "Oh, we forgot about ABC (step, bathroom, camber, terrain)"
Scan an area with ATAP and send it to me - I'll be able to tell whether it's really accessible or not.
1: https://www.google.com/atap/projecttango
Combining crowdsourcing and google street view to identify street-level accessibility problems: http://dl.acm.org/citation.cfm?id=2470744
Thanks for the link!
it is if you want to automate it instead of relying on meatbags.
Your link, just like many other research projects (LA pool database for example) rely on wetware looking at pictures. Depth maps let computers do all the work. You get (well, will get with tango) Quake style maps from simple video clips.
Google did something very smart here. They realized project tango is a long way from shipping, so they released something not perfect, clunky, manually operated, but working _right now_. This way they can claim to be first when new iphone with integrated primesense 3d camera hits the market :).
Can't remember where I read that images to 3D is hard, but this application really gives a good visualization of that problem. It's not the obvious parts that are difficult, but the relatively flat surfaces that are a little distant that cause the big issues.
http://www.theverge.com/2014/4/22/5625264/lytro-changed-phot...
It seems like cool technology - it'd be a shame for it to fall by the wayside because people think software can achieve the same (although if I'm wrong, and it can, please correct me).
But at least my phone is always with me…
I own a 3D printer and hope 3D scanning apps will soon be available.
On the other hand, this demo is showing 3D generation by just filming the object: https://www.youtube.com/watch?v=vEOmzjImsVc
I wonder why this didn't take off.
I think mainly because there isn't much of a usecase for most people.
We use Structure from Motion techniques a lot here where I work (Research Institute) to construct 3D models of buildings and land from UAV imagery.
The current methodologies are great for this use, but they lack the accuracy you'd want to model an object on your desk for 3D printing. That's not to say you can't do it - there's just often a lot of work to be done on the mesh afterwards to make it work.
Autodesk make a pretty good app called 123D catch [0], which uploads your images to their servers for processing.
Alternatively VisualSFM[1] is what I mostly use, I get great results but I don't know how good it is with small things.
[0] http://www.123dapp.com/catch [1] http://ccwu.me/vsfm/