Imagine taking a single photograph of an object and being able to bring that lonely image into editing software that would allow you to turn the object around 360°, revealing all its sides which the camera never even saw. Software that would let you take a single picture (just one frame) of an origami bird and still allow you to turn the bird around, flip it over, even animate it as though you had many photos of the bird from every angle….
Sounds pretty cool, right? Researchers at Carnegie Mellon University thought so, too. The researchers are actually working on a piece software that will allow us to isolate certain objects in our photographs and apply some wicked 3D manipulations to them. And, as it turns out, they are already doing some really cool photo edits with it.
But, how do they do it? Check out this quick video clip for an inside look of the software at work (spoiler alert: the camera doesn’t actually see the unseen…)
Instead of simply editing ‘what we see’ in the photograph, our goal is to manipulate ‘what we know’ about the scene behind the photograph.” — Natasha Kholgade, lead author of the research
The software essentially calls on public databases of objects which contain image data, such as texture, lighting, and dimensions, that match the object from the photo. In the past, this method has not been very successful due to the limited amount of objects found in public databases–the method only works if the object you want to manipulate has a match in the public database. It was also destined to achieve limited success because the image data from the public image would not be a close enough fit, whether it be bad lighting or a be the wrong dimensions, to the original object. To combat this, the researchers developed a method which semi-automatically aligns the image from the public database to the match the geometry of the original object. It then assesses the illumination levels in the original photo and adjusts the public image accordingly.
As for the availability levels of public images, which is, at the moment, not very expansive, the researches say the database is growing by leaps and bounds, even going so far as to suggest that in the very near future, the problem will not be the lack of images in the public database, but rather being able to sort through the mass of images they predict will be added.