FACIAL HAIR SEGMENTATION WITH SYNTHETIC DATA

---

Copy For Citation

Yılmaz Y., Söylemez Pektaş C., Nasiboğlu E.

TASHKENT INTERNATIONAL CONGRESS ON MODERN SCIENCES-III, Toskent, Uzbekistan, 22 - 23 April 2024, pp.1-6

• Publication Type: Conference Paper / Full Text
• City: Toskent
• Country: Uzbekistan
• Page Numbers: pp.1-6
• Dokuz Eylül University Affiliated: Yes

Abstract

 The task of segmenting facial hair from portrait images poses numerous challenges, including the intricate occlusions between facial hair and underlying facial features, and a notable absence of dedicated facial hair mask datasets. To address the challenge of limited real-world data, we unveil a hybrid methodology that combines deep learning with synthetic data generation. We employ StyleGAN [1] to generate a comprehensive dataset, enriching our resources with a wide array of facial hair variations across diverse demographics, thereby overcoming the limitations posed by the lack of real-world data. Our comprehensive experimentation and evaluation demonstrate the network's ability to accurately segment facial hair. The results from extensive quantitative experiments show its effectiveness in segmenting facial hair with high precision. This work represents a significant leap forward in facial image processing, paving the way for future advancements in facial hair analysis and segmentation technologies.