SO(3)-based Continuity Representation to Intensify Head Pose Estimation

In this paper, We proposed an improved succinct structure for head pose estimation from a single image while maintaining continuous representation. which transforms the Head Pose Estimation (HPE) problem into a problem of directly predicting continuous 6D rotation matrix parameters belongs 3D Special Orthogonal Group (SO(3)). The model uses the lighter RepVGG-A2 as the backbone, and the improved Root Mean Square Error of Geodesic Distance(RSME-GD) as the loss function. Experiments show that compared with the original SOTA method, our model parameters are reduced by about 40%, the training speed is 6X faster and the results are even better.