SHARP (Sharp Monocular View Synthesis) is an Apple research model that converts a single photograph into a navigable 3D representation in less than one second. The output is a 3D Gaussian Splat that can be rendered in real time at high photorealistic resolution.
How does it work? Given a single image, SHARP predicts the parameters of a 3DGS representation through a single forward pass of a neural network on a standard GPU. The representation is metric, with absolute scale, enabling realistic camera movements.
The results are impressive: it reduces LPIPS error by 25–34% and DISTS by 21–43% versus the best prior model, while reducing synthesis time by three orders of magnitude.
The project is already available as open-source with a simple CLI:
sharp predict -i /path/to/images -o /path/to/gaussiansOutput .ply files are compatible with multiple public 3DGS renderers. A breakthrough that brings 3D reconstruction to any developer with a single photo and a GPU.
More information at the link 👇
