With the popularity of panoramic X-ray imaging (PXI) and cone beam computed tomography (CBCT) in dental treatment planning and comprehensive prognostic assessment, the demand for effective tooth segmentation is growing. Although 2D panoramic X-rays can effectively detect invisible caries, impacted teeth and supernumerary teeth in children, 3D CBCT has more significant advantages in orthodontic and endodontic treatment, mainly due to its low radiation dose and accurate three-dimensional images. Although 3D images provide more detailed anatomical information, which helps accurate diagnosis and treatment planning, they also face challenges such as difficult data acquisition, radiation exposure, increased processing complexity and metal artifacts. In the field of tooth segmentation, although deep learning methods have made significant progress, the tooth segmentation task still faces many challenges. Compared with tooth semantic segmentation, tooth instance segmentation requires more refined boundary information and accurate capture of tooth morphology, which is particularly difficult in irregular gingival tissue and complex tooth morphology. In addition, the quality of tooth segmentation depends largely on detailed manual annotation, which is not only time-consuming and labor-intensive, but also prone to errors and subjectivity. Recently, STS MICCAI 2024 launched a 3D tooth segmentation challenge, which aims to explore tooth segmentation algorithms based on semi-supervised learning. By participating in this competition, this study explored semi-supervised learning methods through yolov8 [11], aiming to reduce dependence on a large amount of labeled data and train the model by utilizing a large amount of unlabeled data. Finally, among 77 teams, it achieved a score of 0.77 on the validation set, ranking 8th, and a score of 0.77 on the test set.