Tooth instance segmentation is a key technology in the field of medical image segmentation, with applications ranging from orthodontic treatment to dental pathology assessment. Although researchers have developed many tooth instance segmentation models, a common drawback is that traditional techniques are not effective in combining global and local background information and fail to make full use of labeled data, resulting in limited performance in different clinical scenarios. This paper proposes a new deep learning method dedicated to the instance segmentation task of dental panoramic images. By introducing the RFEM attention module and the CBAM attention module into the ResNet architecture, we improve the network’s ability to focus on key features by adaptively assigning weights to different channels and further enhance the fusion effect of global and local features. In order to improve the generalization ability of the data, we use a data augmentation algorithm specifically for labeled datasets. Finally, the network is named the fully supervised tooth segmentation model DAE-Net, which performs well on the MICCAI STS 2024 dataset, with a final performance improvement of 32% compared to the original base model, verifying its effectiveness and stability in the tooth instance segmentation task.