DrFER: Learning Disentangled Representations for 3D Facial Expression Recognition
מושגי ליבה
DrFER introduces disentangled representation learning to enhance 3D facial expression recognition by effectively separating expression and identity information.
תקציר
Facial Expression Recognition (FER) is crucial in understanding human emotions and behaviors.
Early FER methods focused on 2D data, facing challenges like illumination variations.
3D data enables more detailed analysis of facial movements and poses.
DrFER method uses a dual-branch framework to disentangle expression from identity in point cloud data.
Extensive evaluations show DrFER outperforms other 3D FER methods on BU-3DFE and Bosphorus datasets.
The method demonstrates robustness in handling rotational poses.