Sharp
4 days ago
- #Computer Vision
- #Neural Rendering
- #3D Reconstruction
- SHARP is a method for photorealistic view synthesis from a single image.
- It regresses parameters of a 3D Gaussian scene representation in under a second via a neural network.
- The 3D Gaussian representation allows real-time, high-resolution rendering of nearby views.
- The representation is metric, supporting absolute scale and metric camera movements.
- SHARP achieves robust zero-shot generalization across datasets.
- It sets a new state of the art, reducing LPIPS by 25–34% and DISTS by 21–43% compared to prior models.
- Synthesis time is reduced by three orders of magnitude.
- Rendering exceeds 100 frames per second on a standard GPU.
- Demonstrated on photographs from Unsplash.