Symmetry-Based Viewpoint Conversion for Personalized Eye Contact

The learning process is segmented into two distinct phases: a first learning phase that performs viewpoint conversion for unspecified persons using general training data, and a second learning phase that performs personalized learning for specific persons using teacher images and student images. This segmentation allows the system to maintain versatility for general users while achieving high accuracy for specific individuals through targeted personalization.

The system performs preliminary learning in the first learning phase to establish a baseline viewpoint conversion capability for unspecified persons. This preliminary action creates a general-purpose conversion model that ensures versatility. Subsequently, the second learning phase performs personalized adjustments based on teacher images, thereby improving accuracy for specific persons without completely replacing the versatile baseline capability.

The system employs self-service mechanisms where teacher images are automatically captured by the imaging device when the user is in the teaching position, and the learning processor automatically performs personalized learning using these teacher images and corresponding student images. This self-service approach reduces manual intervention and simplifies the complexity of personalized learning deployment.