Apple is developing AI technology that converts ordinary photos into 3D models. The innovation promises fast and accurate results even without specialized equipment.
You might be interested in
Apple In a new research paper, he describes a technology that allows the creation of a realistic 3D model of an object based on just a few ordinary photographs. The method, called “3D Gaussian Splatting with DiT,” uses advanced machine learning and represents a significant advance over previous approaches.
The research documents a system that can generate 3D representations from sources such as images from ordinary iPhone cameras. The goal is to simplify the entire 3D reconstruction process without the need for special equipment or a large number of images from different angles.
What is Gaussian Splatting?
This method uses the concept of "splatting" - spreading three-dimensional data onto a 2D plane, while maintaining the illusion of depth and spatiality. Apple improved this approach by adding a so-called DiT model (Denoising Diffusion Transformer), which optimizes the placement of individual 3D points during learning.
The result is a significantly faster and more accurate conversion of 2D photos into a full-fledged 3D model. According to research Apple Their method comes close to the quality of photogrammetry, but with significantly lower demands on source data and computing power.
Where is that? Apple heading?
Although this technology is not yet commercially available, its potential uses are wide - from improving the function Apple Vision Pro, through augmented reality to advanced tools for content creators or e-commerce. Future use in applications for iPhone or iPad.
Apple has been investing heavily in AI research in recent years, and this work is further evidence that the company does not want to be left behind. The results of this study are publicly available on the arXiv portal.
