Reference PyTorch implementation and models for DINOv3
HFHPE: High-Angle Robust Fast Head Pose Estimation. One-pass yaw/pitch/roll head pose (biternion cos/sin + von Mises κ) distilled from a DINOv3 teacher into compact 64px students, robust to extreme angles and heavy occlusion.
All-in-one training for vision models (YOLO, ViTs, RT-DETR, DINOv3): pretraining, fine-tuning, distillation.