Sharp Monocular View Synthesis in Less Than a Second

lastmod 2025-12-18created 2025-12-182 edits

An approach to synthesizing photorealistic views from a single image

Abstract

We present SHARP, an approach to photorealistic view synthesis from a single image. Given a single photograph, SHARP regresses the parameters of a 3D Gaussian representation of the depicted scene. This is done in less than a second on a standard GPU via a single feedforward pass through a neural network. The 3D Gaussian representation produced by SHARP can then be rendered in real time, yielding high-resolution photorealistic images for nearby views. The representation is metric, with absolute scale, supporting metric camera movements. Experimental results demonstrate that SHARP delivers robust zero-shot generalization across datasets. It sets a new state of the art on multiple datasets, reducing LPIPS by 25–34% and DISTS by 21–43% versus the best prior model, while lowering the synthesis time by three orders of magnitude.

About

• SHARP, released by Apple, is a technique that estimates a 3D Gaussian representation from a single photo to synthesize photorealistic novel views
• Processes in a single neural network feedforward pass in under a second on a standard GPU, and supports real-time rendering
• The generated 3D representation is a metric representation with absolute scale, supporting real camera movement
• Shows zero-shot generalization across multiple datasets, reducing LPIPS by 25–34% and DISTS by 21–43% compared to prior models
• Improves synthesis speed 1000x over prior work, setting a new standard for single-image 3D view synthesis

SHARP overview

• SHARP (Sharp Monocular View Synthesis) is an approach that performs photorealistic 3D view synthesis from a single image
  • Estimates the 3D Gaussian representation parameters of the scene from a single input photo via regression
  • This process completes in under a second on a standard GPU
• The generated 3D Gaussian representation supports real-time rendering and produces high-resolution images from nearby viewpoints
  • Achieves rendering speeds of over 100 frames per second
  • Preserves fine structure and sharp detail

Technical features

• SHARP's 3D representation is a metric representation including absolute scale, reflecting real camera movement
• Processed with only a single feedforward neural network pass, delivering fast results without complex optimization
• Through zero-shot generalization, maintains stable performance even on datasets it was not trained on

Performance and comparison results

• Achieves state-of-the-art performance on multiple datasets
  • 25–34% improvement on the LPIPS metric, 21–43% on the DISTS metric
  • 1000x shorter synthesis time compared to the best prior model
• These gains improve both the efficiency and quality of single-image 3D view synthesis

Visual results

• SHARP visualizes 3D representations generated from single input images, using photos from Unsplash as examples
  • Renderings from nearby viewpoints preserve sharp detail and fine structure
  • Real-time rendering enables natural viewpoint movement

Documentations

[2512.10685] Sharp Monocular View Synthesis in Less Than a Second

https://arxiv.org/abs/2512.10685

See also

• Deep learning

Favorite site

• SHARP web site
• Github - apple/ml-sharp
• SHARP - An approach to synthesizing photorealistic views from a single image | GeekNews