A team of researchers have developed new software that allows users to manipulate a 2D image, by creating a 3D model of the object in questions. Using that model, users can turn, flip, or move the object however they like.
Led by associate robotics researcher Yasser Sheikh, the researchers from Carnegie Mellon developed their experimental software using 3D stock models of common objects: cookware, cars, appliances, and furniture. These models are freely available, and allow the software to provide a reasonable 3D shape from a photo’s 2D image.
“We've created an environment that gives you that same freedom when editing a photo,” says lead author of the study Natasha Kholgade on CNET, who is currently PhD student at the Robotics Institute. “Instead of simply editing 'what we see' in the photograph, our goal is to manipulate 'what we know' about the scene behind the photograph.”
While there are some objects do not have stock models, and while the software does not always match the objects in the photo with an exact 3D copy, it could definitely come in useful for a variety of situations. The team’s software is not limited to 2D objects in digital photography, but can also be used to extrapolate objects from historical paintings or film-based photos.
The researchers programmed the software with semi-automatic alignment—it will automatically shift the 3D model to the shape of the object in the 2D photo—to correct some of its bugs. The software has trouble adjusting an object’s appearance accordingly when presented with a shift in lighting, or may distort a “soft” object because of its imperfections.
The group’s research will be presented at theis year’s SIGGRAPH Conference on Computer Graphics and Interactive Techniques in Vancouver, Canada. If you don’t have time to visit Vancouver, you can read their research here, or visit their website for more information.
Source CNET
Advertisement
