apple-aiml-research/ml-facelit

Official repository of FaceLit: Neural 3D Relightable Faces (CVPR 2023)

What it solves

FaceLit addresses the challenge of creating 3D facial representations that can be realistically relit under different lighting conditions. It allows for the generation of high-quality, 3D-consistent faces that respond naturally to changes in illumination.

How it works

The project uses a neural rendering approach based on EG3D, integrating illumination parameters and camera settings to generate relightable faces. It leverages data preprocessing from DECA to obtain necessary camera and lighting parameters, and supports both diffuse and full lighting modes to handle how light interacts with the face.

Who it’s for

This tool is designed for researchers and developers working in computer vision, 3D face reconstruction, and neural rendering.

Highlights

  • Neural 3D Rendering: Generates 3D-consistent facial images from neural networks.
  • Relighting Capabilities: Supports adjusting lighting conditions to change the appearance of the face.
  • Customizable Control: Allows users to entangle or disentangle specific parameters like camera, light, and specular reflections.
  • Integration: Built upon EG3D and incorporates components from GMPI and DECA.

Related

  • Project
    harry7557558/spirula-studio

    A self-contained 3D Gaussian Splatting pipeline that converts photos and videos into 3D models without requiring Python, PyTorch, or COLMAP.

  • Project
    dweep-desai/FaceGate-Mac

    A native macOS app-locker that uses on-device face recognition, Touch ID, or passwords to restrict access to specific applications.

  • Project
    ros-perception/perception_pcl

    A ROS interface stack that bridges the Point Cloud Library (PCL) with ROS for 3D geometry processing and n-D point cloud handling.

  • Project
    viettranx/3dviz-pro-max

    3Dviz Pro Max is an AI‑agent skill that lets large language models turn a natural‑language description into a complete Three.js scene. It provides a ten‑step workflow, a catalog of 223 recipes, 22 proven Three.js kit blueprints, style defaults, and Python helpers that render the scene in headless Chromium, capture PNG frames, and feed the observations back to the model for refinement. Install by linking the `skills/3dviz-pro-max/` folder into Claude Code or Codex, then run the bundled examples (37 runnable demos) with Vite. The project is MIT‑licensed and focuses on reproducible, evidence‑backed 3D visualisation rather than claiming that every recipe is fully rendered.