Abstract
Minimally invasive glaucoma surgery (MIGS) presents substantial challenges due to its intricate procedures and high technical demands, necessitating robust surgical navigation systems. The development of these systems is highly dependent on the availability of precisely annotated datasets. Addressing this need, we have created the first multicenter, large-scale, fine-annotated surgical video dataset for MIGS. This dataset includes millions of frames, extensively annotated with both surgical instruments and anatomical structures. The annotations are organized into two primary tasks: Task I, which involves phase recognition to identify different stages of the surgery, and Task II, which focuses on semantic segmentation of surgical instruments and anatomical details. By providing a foundational resource for training computer vision models, our dataset aims to facilitate the development of advanced computer-assisted interventions that can improve the precision and safety of glaucoma surgeries.