Last year, Facebook announced a new feature which allowed users to view and post 3D photos. However, Facebook’s 3D photos feature was limited to high-end smartphones that come with Portrait Mode. In order to deliver the 3D photos feature to more users, Facebook has now developed a new state-of-the-art machine learning technique that can convert any 2D photo into a 3D photo. Whether it is a new photo just taken on an Android or iOS device with a standard single camera, or a decades-old image recently uploaded to a phone or laptop, Facebook can convert it into a 3D photo.
Given a standard RGB image, Facebook’s 3D Photos convolutional neural network can estimate a distance from the camera for each pixel. Facebook was able to do it through the following means:
A network architecture built with a set of parameterizable, mobile-optimized neural building blocks.
Automated architecture search to find an effective configuration of these blocks, enabling the system to perform the task in under a second on a wide range of devices.
Quantization-aware training to leverage high-performance INT8 quantization on mobile while minimizing potential quality degradation from the quantization process.
Large amounts of training data derived from public 3D photos.
If you have an iPhone 7 or higher, or a recent midrange or better Android device, you can now try the 3D photos feature in the Facebook app.
You can learn more about how Facebook developed this 3D conversion technique from the source link below.