Aoralscan3 tooth registration dataset

The constructed Aoralscan3 tooth registration dataset includes 1667 samples for training, 156 samples for validation, and 176 samples for testing. Jaw models are generated from hospital patients by oral scanning. The ground truth of the relative pose of each tooth is generated by adding random jittering to the tooth models. For each tooth, ground truth relative pose information was generated by introducing random jittering to the tooth models. This dataset can be used for point cloud registration. which aligns two point clouds and estimates the relative pose between them. In tooth point cloud registration, the inlier ratio (IR) and registration recall (RR) are used as the evaluation criteria. IR measures the fraction of corrected matches (threshold 0.1 mm) in the inferred correspondence set, and it evaluates the accuracy of the predicted correspondence set. RR measures the fraction of correctly registered point cloud pairs, and it is used to evaluate the quality of the predicted 6D pose.