What They Say
IEEE Spectrum published an article about a new AI-based technique that has been developed to generate 3D holograms with very limited processing compared to conventional methods. The full material from the MIT team has been reported in the journal Nature. The method is based on deep learning and is able to generate 60 FullHD 3D holograms with just a single consumer grade GPU and 620K of memory. The group has run it on an iPhone Pro at a rate of 1.1 holograms per second suggesting that we’re not a long way from real time operation.
The system was trained on 4,000 RGB image pairs with depth maps.
There’s an 8 minute video here
Sony sponsored the project.
What We Think
I watched the video and it was interesting, but doesn’t take prisoners! Being able to generate holograms in real time with low power is essential to developing that concept for headsets. (BR)
This image shows different depths to show that the hologram matches the scene.
