maadaa ai

Standard Datasets

Facial 17 Parts Segmentation Dataset

The progress of video and photo editing applications relies heavily on high-quality datasets for machine learning models training. Our carefully selected datasets play a crucial role in enhancing the abilities of such applications, by offering accurate segmentation and recognition of different elements within images and videos.

The "Facial 17 Parts Segmentation Dataset" is specifically compiled for the visual entertainment industry, featuring a range of internet-collected facial images with resolutions exceeding 1024 x 682 pixels. This dataset is dedicated to semantic segmentation, delineating 17 facial categories such as eyebrows, lips, eye pupils, and more. It also includes a selection of portrait images with occlusions, adding complexity and diversity to the dataset for more realistic application scenarios.

Sample

Specification

Dataset ID

MD-Image-075

Dataset Name

Facial 17 Parts Segmentation Dataset

Data Type

Image

Volume

About 2kk

Data Collection

Internet collected images. Resolution is over 1024 x 682.

Annotation

Semantic Segmentation

annotationNotes

Segmentation of 17 categories including left and right eyebrows, upper and lower lips, left and right eye pupils, etc., and a certain percentage of portrait images with occlusion in the image.

Application Scenarios

Visual Entertainment

Data Collections

This dataset consists of approximately 2k images, each segmented into 17 distinct facial categories, providing detailed mapping of facial features. The inclusion of images with occlusions enhances the dataset's utility for developing and testing advanced facial recognition and editing tools, character modeling, and animation systems in visual entertainment. The comprehensive segmentation of facial components allows for nuanced character expressions and more lifelike digital representations in movies, games, and other visual media.

Quality Assurance

Relevant Open Datasets

To supplement our Face Parsing Dataset, users can explore these open datasets for additional resources:

FASSEG Repository [Learn more]

This collection offers datasets for frontal face segmentation (Frontal01 and Frontal02) and a dataset for faces in multiple poses (Multipose01).These datasets can be valuable for training models to perform face segmentation in different orientations and conditions.