It provides a more accurate detection results on the ends and enable realistic hair coloring. Hair texture dataset to train the technology to recognize the hair by its structure.Images taken with a selfie camera to ensure high quality performance on mobile devices.Additionally labelled dataset of over 400 images with the most relevant hair detection results used to assess the hair segmentation accuracy.A variety of settings including backgrounds and low lighting for accurate performance in real-world user environment.'Complex' hair detection tasks to train the neural network effectively on a relatively small initial dataset.Our hair segmentation neural network features: To address these issues, we assembled the balanced hair dataset and perform its ongoing training and improvement. The algorithms trained on studio-like photos will provide low quality resultsВ in real-world user surroundings. Additionally, one needs to take into account the nature of the images. Large hair datasets are hard to obtain, while accurate hair segmentation requires high quality, ground data annotations including different hairstyles, hair lengths, and environmental factors. virtual hair recoloring apps or hairstyle simulators. The algorithm produces a high-quality two-dimensional hair mask extracted from the input image that is well suited for AR applications, e.g.

The core of our hair detection and segmentation technology is a neural network which returns a binary output, tagging the image pixels to human hair or the background. Banuba Hair Segmentation Before-After Demo How does hair segmentation work?

Virtual hair color try on