Three-D Wide Faces (3DWF): Facial Landmark Detection and 3D Reconstruction over a New RGB⁻D Multi-Camera Dataset.

Area of Science:

• Computer Vision
• 3D Imaging
• Machine Learning

Background:

• Deep learning and 3D imaging advancements necessitate comprehensive facial datasets.
• Existing datasets often lack sufficient detail for analyzing facial properties like pose, gender, and age.

Purpose of the Study:

• To introduce the 3D Whole Face (3DWF) dataset, a novel resource for facial analysis.
• To provide a complete dataset with accurate 3D facial data collected via an innovative RGB-D multi-camera setup.

Main Methods:

• Collected 3D raw and registered data for 92 individuals using diverse RGB-D sensing devices.
• Utilized annotated density-normalized 2K clouds and RGB-D streams for data representation.
• Developed an original data augmentation method using face meshes for validation.

Main Results:

• The 3DWF dataset offers accurate visual information for tasks like face tracking and 3D face reconstruction.
• Validated the dataset's reliability through facial landmark detection and head pose classification experiments.
• Demonstrated the alignment and utility of the collected data for machine learning applications.

Conclusions:

• The 3DWF dataset is a valuable resource for advancing research in 3D facial analysis.
• The innovative data collection and validation methods ensure the dataset's high quality and reliability.
• The dataset supports a wide range of facial property analysis tasks, from pose estimation to 3D reconstruction.