Back in 2019, Facebook announced a new feature that will allow users to view and post 3D photos but however, that feature so far Facebook’s 3D photos feature was limited to high-end smartphones exclusively and also came with Portrait Mode.
Facebook has now developed a new state of the art machine learning technique in order to deliver the 3D photo feature to more users that can convert any 2D photo into a 3D photo. Interestingly, Facebook claimed that it won’t matter if the picture is real or new and taken on an Android or an iOS device.
Facebook said that the company’s 3D Photos convolutional neural network can estimate a distance from the camera for each pixel if it is given a standard RGB image. Facebook also said some words on how it is done. They said that network architecture is built with a set of parameterizable, mobile-optimized neural building blocks.
This architecture is an automatic one. It searches to find an effective configuration of these blocks and allows the system to perform the task in under a second on a wide range of devices. Moreover, this architecture was built through quantization aware training to leverage high-performance INT8 quantization on mobile while minimizing potential quality degradation from the quantization process. Facebook has also worked hard to develop it. The company has spent hours behind a large amount of training data derived from public 3D photos.
This means, it does not matter what phone you own, if it is at least an iPhone 7 or higher, or a mid-range Android device, you will be able to try 3D photos feature in the Facebook app.