Face Segmentation

Face Segmentation

Bring to life a variety of face recognition & modification apps by detecting its different parts & features, with image or video processing or in real-time.

Face Segmentation Neural Networks

Our facial feature detection technologies empower you to modify user appearances in augmented reality and face tracking apps. Users can achieve compelling visual effects such as recoloring, maskings, removal, and replacement.

  • Face Segmentation

    The face segmentation features use convolutional neural networks to detect and segment a certain area on the human face including:

    • Full face segmentation
    • Skin segmentation for skin coloring and beautification.
    • Lips segmentation for virtual lipstick try-on.
    • Eye segmentation for iris coloring.
  • Hair Segmentation

    Our hair detection uses deep neural networks to segment the hair in the camera feed. It returns a binary output, tagging the image pixels to the hair or background. You can implement it in beautification or selfie editing apps allowing users to change hairstyles and try on hair colors with a tap. In AR commerce, you can build in-store 'magic mirrors' and hair makeover apps to increase sales, engage consumers, and promote your brand.

Performance

Performance

  • Lightweight neural networks
  • Real-time or post-processing performance
  • Face segmentation models optimized for specific platforms
  • Trained on annotated datasets
  • A wide spectrum of face angles, background settings, and lighting conditions
  • Balanced datasets including nationality, age, and gender 
  • Life-like images to ensure accurate performance in real-world use cases

Neural Networks and Technical Features

  • Background

    • Real-time: 35-40 FPS avg. on mid iOS, Android 
    • Photo: <1 sec. processing time

  • Skin

    • Real-time: 20-30 FPS avg. on mid iOS, Android 
    • Photo: 1-2 sec. processing time

  • Lips

    • Real-time: 18-30 FPS avg. on mid iOS, Android
    • Photo: 1-2 sec. processing time

  • Eyes

    • Real-time: 16-30 FPS avg. on mid iOS, Android 
    • Photo: <1 sec. processing time

  • Hair

    • Real-time: 20-30 FPS avg. on mid iOS, Android 
    • Photo: <1-2 sec. processing time

Why Banuba’s Face Segmentation SDK

Why Banuba’s Face Segmentation SDK

  • We can adapt our technology for custom hardware and use cases.
  • Time and effort save required for algorithm training and testing.
  • Real images and balanced datasets to ensure technology performance for all users
  • We can develop a unique custom technology POC.
  • Patented face segmentation technology
  • Our R&D and PM experts can advise you on the optimal technology to achieve your goals.
FAQ
  • Yes. If you need only one or several features e.g. background separation or hair recolor, we provide you with the custom build and configuration modules that include the features you need.
  • Yes, before purchasing the license you have a 2-week free trial period to validate our SDK performance. To get your free trial period started we need to sign the trial agreement.
  • You can disable Face Tracking manually in any effect so that CPU consumption will be significantly reduced.

    In our tests we reached 62% total CPU usage only with background segmentation enabled

  • A computer vision technology, a subset of image segmentation, dedicated to finding human faces in still pictures and videos. Face segmentation is closely related to the upper body (“selfie”) segmentation but the latter also covers the arms and torso.
  • There are many methods of face recognition and tracking. The most common one is using neural networks that attribute each pixel in the frame to either a person or background. Note that face detection means finding out whether there is a face in the picture. Face segmentation also involves finding its precise location.
  • In computer vision, image segmentation is the process of automatically labeling objects in a picture or video. It is not to be confused with image recognition - finding out whether a specific object is present in the frame. It comes in many varieties: face segmentation, hair segmentation, etc.
  • Simply put, an architectural approach to the process. There are several popular ones (U-Net, Pyramid, etc.) that apply to face segmentation, upper body segmentation, and other use cases.

Get Free Trial