Many children in the 80s will remember various holographic video games that would turn up in malls and arcades. Many of us thought the future would be filled with holographic games and movies, but that hasn’t happened. MIT has developed a new method of generating holograms using artificial intelligence to create holographic content in real-time. The new method is called tensor holography and can be run on a smartphone or laptop.
MIT researchers say that current VR headsets can make users feel sick because VR simulates 3D viewing on a fixed 2D display. Researchers believe a better way to view 3D content could be holograms remade for the digital world. Holograms can deliver excellent representations of the 3D world by shifting the perspective based on the viewer’s position. That fact allows the eye to adjust the focal depth to alternately focus on the foreground and background.
Researchers have been working to make computer-generated holograms, but that process required a supercomputer to run huge amounts of physics simulations. Not only was that time-consuming, but it resulted in less than photorealistic results. MIT’s new method can produce holograms nearly instantaneously. It uses an in-depth learning-based methodology that is efficient enough to run on the laptop or smartphone very quickly.
MIT researcher Liang Shi, the lead researcher on the project, says tensor holography can finally bring the ten-year goal of commercially available holographic displays to reality. Shi also believes the advance could bring holography into fields like VR and 3D printing. Holograms encode the brightness and phase of each lightwave in combination to deliver a more accurate depiction of the parallax and depth of the scene.
Computer-generated holography requires each point in the scene to have a different depth, so the same operations can’t be applied to all of the scene, increasing complexity significantly. Traditional holograms require a clustered supercomputer to run the physics-based simulations and take seconds or minutes to generate a single holographic image. The team’s deep learning methods accelerate computer-generated holography allowing for real-time hologram generation.