Chinese researchers have developed a high-precision three-dimensional (3D) face database and achieved a major breakthrough in personalized facial modeling, a development expected to enhance natural human-computer interaction.
The advancement focuses on improving 3D facial keypoint detection, a core technology required for virtual humans to express emotions, recognize identities, and demonstrate embodied intelligence. Researchers noted that the absence of large-scale, accurately annotated 3D datasets has long limited existing detection algorithms, which often rely on 2D assistance or non-photorealistic models.
A research team from the Shenzhen Institutes of Advanced Technology of the Chinese Academy of Sciences and Fujian University of Technology developed a new curvature-fused graph attention network (CF-GAT). The system can predict facial landmarks directly from raw point clouds, enabling a shift from “one-size-fits-all” approaches to personalized modeling.
The study was published in the journal IEEE Transactions on Circuits and Systems for Video Technology.
The team also constructed a customized 3D/4D facial acquisition system and completed standardized data collection, creating what it described as the industry’s largest high-precision, accurately annotated 3D facial database. The database contains approximately 200,000 high-fidelity 3D facial scans.
In addition, the system includes a multi-expression 3D face dataset, a standardized 3D facial landmark dataset, a high-precision 3D human body dataset, and a dynamic 4D facial expression dataset.
According to corresponding author Song Zhan, these databases provide foundational support for key technologies in humanoid robotics, enabling high-fidelity perception, expression modeling and behavior generation.
He added that the datasets will further support data-driven large-model humanoid robot systems, helping build more natural and intelligent human-robot interaction capabilities in the future.
