MIT IBM Train AI to Create and Edit Fake Images

first_imgStay on target Researchers from MIT and IBM trained artificial intelligence to generate realistic photographic images, then edit objects inside them.Dubbed GANpaint Studio, the system would be an ideal assistant to artists and designers looking to make quick adjustments to visuals.But it also offers insight into how neural networks learn context, and could help computer scientists identify fake or altered content.Take GANpaint Studio for a test drive with the interactive demo.Simply upload an image and choose which aspects to modify: transform a wall into a window, a dome into a spire, a green spring tree into orange autumnal foliage.Spearheaded by MIT professor Antonio Torralba, as part of the MIT-IBM Watson AI Lab he directs, the project could be used to improve or debug other GANs (generative adversarial networks) under development.“Right now, machine learning systems are these black boxes that we don’t always know how to improve, kind of like those old TV sets that you have to fix by hitting them on the side,” according to David Bau, a PhD student at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL).“This research suggests that, while it might be scary to open up the TV and take a look at all the wires, there’s going to be a lot of meaningful information in there,” lead study author Bau said in a statement.One unexpected discovery was that the system has learned simple rules about the relationships between objects. It knows not to put something where it doesn’t belong—like a window in the sky—or to insert different doors on two different buildings.“All drawing apps will follow user instructions, but ours might decide not to draw anything if the user commands to put an object in an impossible location,” Torralba said.The team’s goal is to give people more control over GANs, and ultimately better tools to stamp out forgeries.“You need to know your opponent before you can defend against it,” co-author Jun-Yan Zhu, a postdoc at CSAIL, said. “This understanding may potentially help us detect fake images more easily.”Adobe recently trained a convolutional neural network (CNN) to identify altered images of faces. The tool also pinpointed specific areas and methods of facial warping, and was able to revert images to what it estimated was their original state.The results, according to Adobe, impressed “even the researchers.”More on Robot Learns to ID Objects by Sight, TouchNew AI Tool Can Help Doctors Detect Brain AneurysmsMIT Robot Helps Lit Objects by Looking at Your Biceps McDonald’s Plans to Serve AI Voice Technology at Drive ThruCIMON Returns to Earth After 14 Months on ISS last_img

Leave a Reply

Your email address will not be published. Required fields are marked *