AI for Real-Time Holograms

With a new method called tensor holography, holograms can be created in virtual reality, 3D printing, medical imaging and many other fields, and this method can work on smartphones. Although VR technology is developing day by day, it has not yet been able to overthrow TVs or computer screens as devices used to watch videos. The main reason for this is that VR can make people sick. Although users in VR technology are actually looking at a 2D fixed screen, this screen creates a 3D impression for the user. This situation can cause nausea and eye strain in the person during long-term use. The solution to this situation is a 60-year-old technology restructured for the digital world: hologram . Holograms offer an extraordinary representation of the 3D world around us. Holograms provide a perspective that changes with the viewer's position and allow the eye to adjust its focal depth to alternately focus on the foreground and background. Researchers have long been trying to create computer-generated holograms, but this process was time-consuming and yielded unrealistic results. Researchers now MIT say that researchers have developed a new way to produce holograms almost instantly, and the deep learning-based method is efficient enough to run on a laptop in the blink of an eye. "People previously thought it was impossible to do real-time 3D holography calculations with existing consumer-grade hardware," says Liang Shi, lead author of the study and a PhD student in the Department of Electrical Engineering and Computer Science (EECS) at MIT. Shi believes that the new approach, which the team calls "tensor holography," will eventually make this goal attainable. The advance could trigger the spread of holograms to areas such as virtual reality and 3D printing. Researchers used deep learning to accelerate computer-generated holography and allowed real-time hologram generation. The team designed a convolutional neural network to roughly mimic how humans process visual information. Training a neural network usually requires a large, high-quality dataset that was not previously available for 3D holograms. The team created a special database of 4,000 pairs of computer-generated images. Each pair matched images, including color and depth information for each pixel, with its corresponding hologram. In the new database, researchers used scenes with complex and variable shapes and colors to create holograms, pixel depth was evenly distributed from foreground to background, and they implemented a new physics-based computational set to eliminate occlusion. This approach resulted in photorealistic training data. They then revealed that their algorithm achieved a successful result. Learning from each image pair, the tensor network changed the parameters of its own calculations and repeatedly increased its ability to create holograms.   This article is cited.