Repolode
Submit
apple-aiml-research

apple-aiml-research/ml-sharp

Apple's research code that turns a single photo into a 3D Gaussian scene in under a second, renderable as nearby novel views.

View on GitHubWebsite
Stars
8,989
Forks
652
Open issues
76
Last commit
last month

What is it?

SHARP is a research project from Apple that performs monocular view synthesis: given one photograph, it predicts the parameters of a 3D Gaussian representation of the depicted scene in a single feedforward pass of a neural network. On a standard GPU this takes less than a second.

The resulting Gaussian scene can be rendered in real time to produce high-resolution images from nearby camera viewpoints. The representation is metric, with absolute scale, so camera movements are expressed in real units. Output is written as 3DGS .ply files, which the project notes are compatible with various public 3DGS renderers, using the OpenCV coordinate convention.

The project provides a `sharp` command-line tool with `predict` and `render` subcommands. Prediction runs on CPU, CUDA, or MPS, while rendering camera trajectory videos currently requires a CUDA GPU. The model checkpoint downloads automatically on first run.

Key features

  • 01Generates a 3D Gaussian scene from a single image in one feedforward pass.
  • 02Runs in under a second on a standard GPU.
  • 03Produces metric-scale representations that support real-unit camera movement.
  • 04Exports standard 3DGS .ply files compatible with public renderers.
  • 05Renders camera trajectory videos in real time (CUDA GPU required).

Use cases

  • Turning a single photo into an explorable 3D scene for nearby viewpoints.
  • Prototyping 3D Gaussian splatting pipelines without multi-view capture.
  • Generating camera fly-through videos from still images.

Quick start

  1. 1

    Create a Python environment

    conda create -n sharp python=3.13
  2. 2

    Install the project dependencies

    pip install -r requirements.txt
  3. 3

    Test the installation

    sharp --help
  4. 4

    Run prediction on your input images (the checkpoint downloads automatically on first run)

    sharp predict -i /path/to/input/images -o /path/to/output/gaussians

Things to know

  • Rendering videos via the --render option requires a CUDA GPU.
  • The README does not state a software license; check the LICENSE and LICENSE_MODEL files before use.

Topics

#computer-vision#view-synthesis#gaussian-splatting#3d#deep-learning#python

FAQ

What is ml-sharp?
Apple's research code that turns a single photo into a 3D Gaussian scene in under a second, renderable as nearby novel views. SHARP is a research project from Apple that performs monocular view synthesis: given one photograph, it predicts the parameters of a 3D Gaussian representation of the depicted scene in a single feedforward pass of a neural network. On a standard GPU this takes less than a second.
Is ml-sharp free to use? What license does it have?
ml-sharp is open source under the Other license. Check the license terms before using it in commercial products.
Is ml-sharp still maintained?
Updated within 6 months. The last commit to ml-sharp was on Sep 11, 2026.
What can I use ml-sharp for?
Turning a single photo into an explorable 3D scene for nearby viewpoints. Prototyping 3D Gaussian splatting pipelines without multi-view capture. Generating camera fly-through videos from still images.
How do I install ml-sharp?
Create a Python environment: conda create -n sharp python=3.13 Install the project dependencies: pip install -r requirements.txt Test the installation: sharp --help Run prediction on your input images (the checkpoint downloads automatically on first run): sharp predict -i /path/to/input/images -o /path/to/output/gaussians
Where can I get ml-sharp?
The source code and installation instructions are on GitHub: https://github.com/apple-aiml-research/ml-sharp

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects