In the field of 3D content creation, Apple recently dropped a "deep-sea bomb." According to tech media 9to5Mac, Apple has officially open-sourced a new AI model called SHARP. The most astonishing aspect of this technology is that it breaks through the traditional time-consuming bottleneck of 3D reconstruction, which used to take hours, and can convert a regular 2D photo into a 3D scene with real physical proportions in less than a second.
According to Apple's latest paper titled "One Second to Achieve Clear Monocular View Synthesis," the core secret of SHARP lies in its advanced "3D Gaussian Splatting" technology. Unlike previous complex processes that required taking photos from hundreds or even thousands of angles, SHARP has mastered general spatial geometric rules through deep training on massive data. This means it can directly predict the positions of millions of "Gaussian spheres" with lighting information through a single fast "scan" by a neural network, completing modeling instantly.
In terms of image quality, SHARP also sets a new industry benchmark. Test data shows that the 3D views generated by it significantly outperform the strongest models in the industry in terms of texture details and structural accuracy, and can support extremely realistic camera movement simulations.
Currently, Apple has released the complete code and resources of SHARP on the GitHub platform for global developers to download. Although the model currently focuses mainly on reconstruction near the original image perspective and cannot fully "imagine" the blind spots behind, its nearly real-time conversion experience will undoubtedly open up new possibilities for mobile 3D creation and spatial computing applications.
Key points:
⚡ Speed breakthrough: The SHARP model has improved the processing speed of 2D to 3D conversion by three orders of magnitude, achieving an almost real-time conversion experience in less than a second.
🌐 Leading 3D generation technology: Based on 3D Gaussian Splatting technology, the model can predict millions of 3D point positions through a single neural network forward pass, accurately restoring real-world physical proportions.
🔓 Comprehensive open-source ecosystem: Apple has open-sourced the code and resources of SHARP on GitHub, aiming to accelerate innovation in spatial computing and 3D content creation worldwide.
