Apple has just unveiled a groundbreaking technology in the AI field of 3D reconstruction, which has long been considered a "tough nut" to crack.
According to the latest report, Apple's AI research team has released a new model called LiTo (Surface Light Field Tokenization). Its core breakthrough lies in:reconstructing a complete 3D object from just a regular 2D image, with a level of detail that reaches physical realism.
For a long time, the biggest challenge in generating 3D models from a single image has been "lighting consistency." When you change the viewing angle, the surface reflections and highlights on the object often become distorted or unrealistic. However, the LiTo model has successfully solved this problem by introducing an innovative "latent space" representation. Instead of simply memorizing pixels, it now understands the fundamental laws of how light interacts with surfaces through mathematical vectors.
In simple terms, LiTo has strong "imagination" capabilities. Even with just a front-facing photo, it can accurately predict mirror highlights and Fresnel reflections on the back of the object under different lighting conditions. In official comparison tests, LiTo has significantly surpassed the industry-leading TRELLIS model in multi-view lighting accuracy.
To refine this "detail-oriented" AI, researchers trained it using thousands of 3D objects across 150 perspectives and three lighting conditions. This almost obsessive pursuit of lighting details clearly indicates Apple's efforts to lay the groundwork for a spatial computing ecosystem.
Imagine, in the future, you only need to take a photo with your iPhone, and LiTo will instantly convert it into a lifelike 3D model, seamlessly placed into the Vision Pro virtual space. This seamless transition from 2D content to 3D assets might be the key weapon for Apple to achieve a "late-mover advantage" in the AI race.
