Macular buckling is the primary treatment for myopic foveoschisis, requiring implants individually designed to each patient’s ocular morphology. And the inferior oblique muscle is a key anatomical landmark for determining implant suture points. However, there is not any dataset with inferior oblique muscle annotation. And the existing ocular region image segmentation methods struggle to focus on the small and irregular inferior oblique muscle. In this paper, we construct a dataset containing labels for inferior oblique muscle and propose an anatomical prior guided progressive inferior oblique muscle segmentation method. Furthermore, we employ a semi-supervised learning strategy to generate pseudo-labels in nnU-Net model, which achieves high performance for both eyeball and inferior oblique muscle segmentation with few annotated CT.