EasyPortrait: Face Parsing and Portrait Segmentation Dataset

Apps with video calls have grown in popularity in recent years. Many use them daily for work, school, or to keep in touch with friends and family. Therefore, the functionality of video conferencing software strongly began to increase, adding new features based on Machine Learning models to their ecosystem. Background removal or its blur is one of the most used features in this type of software. It’s also worth noting Face Beautification appears in them, which implies blurring or lightening the face skin, whitening teeth, and applying makeup effects. But as you can see for yourself, the quality of these features work in some cases is unsatisfactory and needs to be improved. One of the reasons for this problem is the lack of data for this domain, which includes some features, such as the presence of headphones on the person or the background close to the person. Most of the Portrait Segmentation datasets were marked up automatically, using a green screen or Adobe Photoshop, indicating they are not high quality. The rest are marked up manually, have low heterogeneity in subjects and scenes, or don’t fit in the domain of video conference images. But the situation is more complicated with Face Parsing datasets. Most contain a small number of samples for training, low-quality photos, and segmentation masks with questionable annotation. <a href="https://betterprogramming.pub/easyportrait-face-parsing-and-portrait-segmentation-dataset-bb38eaf1b082">Website</a>