Apple just dropped another AI flex. Their researchers built a model named SHARP that can spin a single 2D photo into a full 3D scene in under a second. This thing runs on a standard GPU, spitting out a photorealistic result from one picture by guessing what the nearby viewpoints would look like.
The model uses a technique called 3D Gaussian Splatting, which normally needs a bunch of images from different angles to build a scene from millions of tiny colored blobs. SHARP bypasses that need, predicting depth and color from a solitary image to create a navigable, metric scale environment you can render in real time.
It is another sign of Apple pouring gas on its AI development, pushing speed and practicality. The whole point is making complex scene generation almost instantaneous, a move that could seriously change things for graphics and AR down the line.
The model uses a technique called 3D Gaussian Splatting, which normally needs a bunch of images from different angles to build a scene from millions of tiny colored blobs. SHARP bypasses that need, predicting depth and color from a solitary image to create a navigable, metric scale environment you can render in real time.
It is another sign of Apple pouring gas on its AI development, pushing speed and practicality. The whole point is making complex scene generation almost instantaneous, a move that could seriously change things for graphics and AR down the line.
