CelebHair - Face/hair segmentations 


Automatic hair (style) analysis has implications in various domains: soft biometrics, fashion, visagisme, avatar generation, just to name a few. However, there are few datasets which provide annotations and images for this task. The purpose of this dataset is to provide segmentation masks (labeled with face, hair and background pixels) for more than 3500 unconstrained, "in-the-wild" face images. The input images are taken from the CelebA . We do not own the input images so you have to contact the authors to obtain permission to use the corresponding input images. 

The segemntation masks correspond to the aligned and cropped png images from the CelebA dataset. Each mask is a bmp file with the same basename as its corresponding input image (for example, the segmentation mask 000567.bmp corresponds to the image 000567.png).

The segmentation masks contain only the following pixels 0 (background pixel), 128 (face area pixel) or 255 (hair area pixel).




Please fill in the agreeement form in order to receive the data!

The database contains the following files:

Terms and conditions


If you use these data please cite the following work: 

Borza, Diana, Tudor Ileni, and Adrian Darabant. "A Deep Learning Approach to Hair Segmentation and Color Extraction from Facial Images." International Conference on Advanced Concepts for Intelligent Vision Systems. Springer, Cham, 2018.

Bibtex format:

  title={A Deep Learning Approach to Hair Segmentation and Color Extraction from Facial Images},
  author={Borza, Diana and Ileni, Tudor and Darabant, Adrian},
  booktitle={International Conference on Advanced Concepts for Intelligent Vision Systems},

Related datasets

The input images for the annotation masks are obtained from the CelebA (Large-scale CelebFaces Attributes Dataset).

Other databases with facial hair segmentation or analysis:



Please contact dadi (at) cs.ubbcluj.ro Adrian DARABANT or Diana Borza (diana.borza (at) cs.utcluj.ro)  or Tudor Ileni (ileni.tudor (at) gmail.com) for any questions or comments about the database.