yakhyo/face-parsing

Real-time face parsing and facial semantic segmentation with BiSeNet - PyTorch training, ONNX export, pretrained weights.

What it solves

This project provides a precise way to perform "face parsing," which is the process of automatically identifying and segmenting specific facial features—such as eyes, nose, mouth, and skin—within an image. This technology is essential for applications like digital makeup, augmented reality (AR) filters, facial recognition, and emotion detection.

How it works

It uses a "Bilateral Segmentation Network" (BiSeNet) architecture. The model takes a face image as input and produces a detailed mask that maps each pixel to a specific facial component. You can choose between two backbone architectures, ResNet18 or ResNet34, depending on your needs for speed or precision. The repository includes tools to train your own models, run inference on images, and export the models to the ONNX format for use in other software environments.

Who it’s for

This tool is designed for developers and researchers working on computer vision projects who need a reliable, pre-trained, or customizable model for facial feature extraction. It is particularly useful for those building AR applications or automated image editing tools.

Highlights

  • Highly accurate segmentation of facial landmarks and components.
  • Supports both ResNet18 and ResNet34 backbone architectures.
  • Includes full pipelines for training, inference, and ONNX model conversion.
  • Pre-trained weights are available for immediate use.
  • Proven reliability, as it is utilized by major open-source projects like facefusion.

Related

  • Project
    aaryansamanta/advanced-ai-curriculum

    This repository showcases original capstone projects and coursework from advanced AI and Data Science programs, including CNN-based pneumonia detection, diabetes prediction, and healthcare data analysis. It emphasizes applied deep learning and machine learning for healthcare while protecting unpublished research.

  • Project
    ncoevoet/facet

    Facet is a private, self-hosted photo manager that scores each image across 9 quality dimensions, then offers a web gallery with culling, face recognition, semantic search, sharing links, and optional GPU-accelerated AI — all offline with no cloud accounts.

  • Project
    Sompote/DINOV3-YOLOV12

    A hybrid object detection framework combining YOLOv12 and DINOv3 to provide superior accuracy and faster convergence, especially on small or complex datasets.

  • Project
    choosehappy/HistoQC

    HistoQC is an open-source quality control tool for digital pathology slides, helping users assess and manage the quality of whole slide images before downstream analysis. It is backed by peer-reviewed research and provides a standardized approach for pathology workflows.