Introducing Digg Dialogg!
Check out the first Digg Dialogg with Nancy Pelosi. More guests to be announced soon!
Turn Any Picture Into 3D Image
techcrunch.com — Fotowoosh, a new service from Maryland-based startup Freewebs, will turn any image into a 3D model. In a week or so, the company says, users will be able to upload a picture and have an animated 3D image returned to them in an embeddable Flash widget.
webtech- 2180 diggs
- digg it
- jstohler, on 10/12/2007, -10/+54That's badass.
- geekme, on 10/12/2007, -29/+14Cool, wonder how it works?
- W00DR0W, on 10/12/2007, -9/+59@geekme
It involves a mouse and a big wheel. - ChicknBot, on 10/12/2007, -31/+7I wonder how this could be applied to video games to reduce GPU usage, specially on FPS games.
- edilclyde, on 10/12/2007, -10/+109Goatse in 3D
- zizzy, on 10/12/2007, -8/+3@geekme
I'm guessing you didn't read the article... - edilclyde, on 10/12/2007, -1/+24for those wondering like geekme (from the download README )
README for Automatic Photo Pop-up
The photo pop-up algorithm consists of two main stages: estimating a rough labeling of the geometry of the image and using that geometric information to construct a 3D model. The first of these stages tends to be fairly accurate and results in probability maps for each of three main geometric classes (ground, vertical, sky) and five subclasses of vertical (planar facing left/right/center, porous, and solid). The second stage of constructing the model works in limited scenarios. If you can imagine constructing a pop-up of the image by making a few straight cuts and a few straight folds, then the algorithm might be successful; otherwise it is likely to give some very rough approximation of the scene.
Details on how our method works can be found in our SIGGRAPH, ICCV, and IJCV (draft) papers online at www.cs.cmu.edu/~dhoiem/publications/. This online version is from the ICCV paper, which may produce slightly different results than the SIGGRAPH or IJCV versions. Also, because the algorithm is random, it is possible to get slightly different results on different trials of the same image.
Input:
One color image. The algorithm has been trained on real outdoor images taken from the ground, but it is sometimes fun to see if it works on paintings, indoor images, and strange viewpoints.
Output:
The primary output of the algorithm is a labeling into geometric classes with associated probability maps:
*.l.jpg: a labeling of the image (green=support, red=vertical, blue=sky ; arrows indicate planar directions, 'O' for porous, 'X' for solid)
*.v000.pgm: probability of support (e.g. ground) given image
*.v090.pgm: probability of vertical (i.e., anything sticking out of the ground) given image
*.vsky.pgm: probability of sky given image
*.h045.pgm: probability of planar (e.g. wall), facing left given vertical, image
*.h090.pgm: probability of planar, facing center given vertical, image
*.h135.pgm: probability of planar, facing right given vertical, image
*.hpor.pgm: probability of porous surface (e.g. tree leaves) given vertical, image
*.hsol.pgm: probability of solid non-planar surface (e.g. a person) given vertical, image
The secondary output of the algorithm is the photo pop-up model:
*.wrl: VRML model (requires VRML viewer such as Cortona to view)
*.png: texture maps for 3D model
Yup I dont get it either... - airwalkery2k, on 10/12/2007, -8/+4"Goatse? What's that?"
Eww in 3 dimensions now! - Aliarse, on 10/12/2007, -4/+3@edilclyde - Thats just wrong, is the 2d not bad enough for you!?
- robbh66, on 10/12/2007, -5/+5Pfft, VRML.
I took that class for some god-awful reason in college. It sucked then, it sucks now. It's not going anywhere. - maiku00, on 10/12/2007, -3/+4VERY cool. You could do this to some pretty neat things. I wonder what some of the 9/11 Photos would look like?
- ayeroxor, on 10/12/2007, -5/+2geekme: Question marks are usually reserved for questions.
- nevesis, on 10/12/2007, -6/+96Am I the only person who read this and thought "porn" ?
- Oly701, on 10/12/2007, -0/+22Ha, surely you jest.
Every inventor and entrepreneur knows the easiest way to a successful product is make it somehow useful to the porn industry (and of course, don't get involved in land wars in Asia). - hdtvdust, on 10/12/2007, -17/+3Oly...other than VHS, what product or service benefitted greatly from the porn industry?
- Aliarse, on 10/12/2007, -1/+23Silicone.
- troymcdavis, on 10/12/2007, -0/+15@ hdtvdust
Can anything else explain the ubiquity of high-speed internet? - Heembo, on 10/12/2007, -1/+7hdtvdust : The gutenberg printing press and every other media device since then know to man, you idiot!
- rrasco, on 10/12/2007, -2/+2and whats the cost on something like that? probably worth more than its awe. still pretty neat concept.
- Oly701, on 10/12/2007, -0/+22Ha, surely you jest.
- evanstapler, on 10/12/2007, -4/+43This brings pornography into a whole new DIMMENSION!
- Supurcell, on 10/12/2007, -1/+146I've always wondered what the naked female form looked like in three dimensions.
- Wezlanator, on 10/12/2007, -11/+31*takes picture of penis*
- Recuso, on 10/12/2007, -1/+190Unfortunately the object has to originally have some height/depth to it.
- dpoehls, on 10/12/2007, -0/+1@recurso
Ha!
- niradg, on 10/12/2007, -2/+18people still use VRML?
- cenithx, on 10/12/2007, -1/+7I was just going to comment.. I thought VRML was dead? When it first came out I played with it for a bit, relatively excited about it.. but it seemed to fizzle out and vanish completely..
- Flashman, on 10/12/2007, -0/+8"The 3D image is constructed in Virtual Reality Modeling Language (VRML) format, meaning you currently need a VRML reader to see it (future browsers will likely build this functionality in)."
Hahaha - 13 years late, I think. - dacheetah, on 10/12/2007, -0/+5lol yeah. I remember playing with VRML back in early high school a decade ago, when they were saying it'll be implemented into browers as a default soon.
It's not a bad markup language, but it's also basically dead, and I don't think we'll see it as a default option in many browsers. - omeomi, on 10/12/2007, -1/+1I think it's funny that the article says "you currently need a VRML reader to see it (future browsers will likely build this functionality in)". Future browsers? Netscape 4 would parse VRML...if current browsers don't, it's because VRML sucks, and nobody wanted it...
- rebelxtnut, on 10/12/2007, -8/+3PORN hehe this is gona be so sweet
- androothebear, on 10/12/2007, -6/+3where do i download?
- Scheissen, on 10/12/2007, -6/+1That's old.
- zizzy, on 10/12/2007, -0/+5A bunch of examples:
http://www.freewebs.com/fotowoosh/demo.html - gheide, on 10/12/2007, -5/+1How the heck are they doing this? I don't see how it can be completely automated. Lets say you take 3 pictures of the same corner of a building, just different vertical horizon angles. Does their system reliably put the image around the corner of a box representing the building precisely in all 3 images? I'd like to see more of how and how fast it actually works... without any distance calculations external of the image, how do you get a 3d representation? Each pixel doesn't get distance info embedded in EXIF data - yet...
- op12, on 10/12/2007, -3/+1You can use lighting and angles to guess the depth, though I don't think this will use multiple images for information as I think you're suggesting. I'm interested to see if it will be able to remove distortion when you view the 3d object from a different angle. From the sample video on their homepage right now, it looks like it can only minimally change perspective, but if you try to look from a much different angle, the object will appear significantly distorted (which is not really surprising...there's a pretty limited amount of 3D information as you said).
- quickgold192, on 10/12/2007, -3/+1think about it, when you look at a picture with your own eyes, your brain can come up with a pretty good guess as to what it would be like in 3D.
- LordLandon, on 10/12/2007, -0/+1@quickgold
when you look at something with your eyes, you have two perspectives, due to two eyes. your brain uses the differences between the two perspectives to estimate depth
in the case of a picture, you only have one perspective and depth cannot be estimated as easily. - nodnarb24, on 10/12/2007, -0/+2If you are interested in how they did it, you can read their paper on it here: http://www.cs.cmu.edu/~dhoiem/publications/popup.pdf
It can work on any image but only really works well on outdoor images, especially of man-made structures. The paper shows some of their failure images as well.
- chadma3, on 10/12/2007, -8/+0This is the coolest thing ever it will change the movie industry, the real estate industry and many more
-----------------------
Ihatebull.com - Hamsterpotpies, on 10/12/2007, -3/+1Even if this is old news, the site keeps getting better and better!
- Kyderdog, on 10/12/2007, -8/+1Isn't this just a modified Ken Burns Effect?
Nothing Really new here its been in iLife forever...- grogan, on 10/12/2007, -1/+3Did you actually read the article? This is using heuristics to make a 3D model of the scene depicted by a photo. The zooming and panning is just a way to show that off.
From http://en.wikipedia.org/wiki/Ken_Burns_Effect :
The Ken Burns Effect, made famous by American documentarian Ken Burns, refers to a technique of embedding still photographs in motion pictures, displayed with slow zooming and panning effects, and fading transitions between them.
- grogan, on 10/12/2007, -1/+3Did you actually read the article? This is using heuristics to make a 3D model of the scene depicted by a photo. The zooming and panning is just a way to show that off.
- Dan11023, on 10/12/2007, -2/+1i wish they used this technology in video games
- freonchill, on 10/12/2007, -6/+3this DOES NOT WORK
i tried it when it was posted on digg the first time (or was it just last time and its been on there 3-4 times by now
i downloaded all the files it said i needed
and even used their test image to test it out
didnt work
tried a simple image of my own - that i made in sketchup that was cubes
didnt work
and by didnt work - i mean the program didnt finish running and errored out - not that the images looked like they were in 3D
oh well, so much for that - jongos, on 10/12/2007, -1/+1I'm sure this works, but will it work well enough to warrant the 11 million dollars in venture capital they've obtained?
- shervster, on 10/12/2007, -0/+1remember this works from just one single picture. the algorithm for indoor pictures is coming. we will widgetize this so people can view this through flash and publish and share their 3D pictures anywhere without ever using VRML. think about the cool things you could do with this if we mash it up with google maps or google earth and allow you to see the street scenes at the ground level from single pictures taken at the ground level. mathematically this was impossible. it really takes cues from how the human brain infers a scene and sees it in 3D.
- swordedge, on 10/12/2007, -0/+2seen something like this before. Not really 3D, just fancy processing of a 2D picture. Still nice though
- pbgswd, on 10/12/2007, -1/+3vrml is dead. I really loved stuff I saw in vrml, but its dead dead dead now.
- ConceptJunkie, on 10/12/2007, -0/+1Really? I never saw anything in VRML that didn't look like state-of-the-art 3D from 1993, i.e., like crap.
- ConceptJunkie, on 10/12/2007, -0/+1Really? I never saw anything in VRML that didn't look like state-of-the-art 3D from 1993, i.e., like crap.
- keithburgun, on 10/12/2007, -1/+0awesome
- kutza, on 10/12/2007, -0/+1Hrmm... Trying to install the cortata plugin on firefox in vista, but it's giving me an error 202 (access denied). Anyone have any clue to bypass this?
- shortarabguy, on 10/12/2007, -0/+1It looks like this has been having trouble from the start... Is this thing even working?
- digitalArtform, on 10/12/2007, -0/+1I do mine like this:
http://www.digitalartform.com/archives/2004/11/camera_projecti_1.html- kohan69, on 10/12/2007, -0/+1@maya 3d mapping now THAT is really cool.
This 3D picture thing would be useful if it at least exported a picture into a 3ds model + texture
- kohan69, on 10/12/2007, -0/+1@maya 3d mapping now THAT is really cool.
- shortarabguy, on 10/12/2007, -1/+1This seems a lot like something that Google would be acquiring...
Don't be surprised if Google comes out with some implementation of a similar technology for Google Earth. Don't expect it in Google Maps though... There's relatively little use for it when you're just getting directions. - mash8591, on 10/12/2007, -0/+7I swear I have seen that train and that house somwhere before...
- moft, on 10/12/2007, -3/+1wow - oooold technology.
this sort of ***** has been around for ages. i guess this digg entry is for those who don't know.- morphie, on 10/12/2007, -0/+2So, come up with a link older than a year then. We all know this have been on digg before. The thing is, now it becomes available to all online.
- mookieXL, on 10/12/2007, -0/+1moft is right. Canoma (and few other programs) can do same thing and results look much better. (But it requires some manual editing)
- harrymcwealth, on 10/12/2007, -0/+1When I first saw it, I thought it would fill in the unknown area behind (too). You know kind of like a human. If you cover half a letter, they figure out what the rest is. Now that would be amazing.
- ipodman715, on 10/12/2007, -1/+2Wow, something from Maryland! Go Maryland! :P
- TimDigg, on 10/12/2007, -0/+2Funny because I was just about to move to Cali....ironic because someone was just telling me yesterday "you underestimate the talents in maryland"....
- TimDigg, on 10/12/2007, -0/+2Funny because I was just about to move to Cali....ironic because someone was just telling me yesterday "you underestimate the talents in maryland"....
- TimDigg, on 10/12/2007, -1/+2Dugg for "Maryland-Based"
- b04155, on 10/12/2007, -0/+1I'm just as impressed at the last time I saw this, ie no improvement. It's not a dupe I'm referring to rather the work finally moved from the lab into a real world app. However, they're still demo-ing the SAME pictures that were seen last time.
- mixd, on 10/12/2007, -0/+1The Bay, crabs, football, and now... web-based 2D to 3D image processing.
- motters, on 10/12/2007, -0/+1I know the inner details of how this system works, and it's really not as impressive as the animations might suggest. The 3D photo-popup effect is *very* approximate and only works - to a degree - under certain limiting circumstances. It certainly doesn't work with just any old photo. If I was a venture capitalist interested in 3D photo technology I'd put my money into something like photosynth rather than the CMU photo-popup system.
- nick2, on 10/12/2007, -0/+1This looks awesome!
- cavar, on 10/12/2007, -0/+0I can do this with Canoma.
- Kitsune818, on 10/12/2007, -0/+1This isn't really 3D at all, its more of a perspective trick. The farther you get from the original point of view, the more the objects will be distorted. Imagine if you could move into that picture of the train a bit, then turn and look back.. the texture on the 3D model would have the angles all wrong.
Remember that guy who draws "3D" images on the sidewalk with chalk? But it only works if you look at it just right, and with only one eye. If you look at it from any other angle it's hard to tell what it's supposed to be. It also looks really impressive in photos because there is only a single POV, but with two human eyes in person it doesn't look nearly 3D at all (Although it's pretty cool).
This is just a passing thing, there isn't enough information there to do much more than those little panning demos they've already shown us. - Cyberdactyl, on 10/12/2007, -0/+1Google Sketchup has had that as an imbedded function called photomatch, then simply exporting a avi.
- FastZ, on 10/12/2007, -0/+1Why must they delete the sky in each of the pictures? Can they not use that in the 3d version?
Digg is coming to a city (and computer) near you! Check out all the details on our 
© Digg Inc. 2008 —
Content posted by