Synthetic Monocular Depth Estimation Dataset for Cataract Surgery Assistance

In computer-assisted surgeries, monocular depth estimation plays an important role, which provides navigation for surgeons by computing precise depth information. In recent years, depth estimation has achieved significant breakthroughs with the application of deep learning. However, the lack of depth ground truth in the ophthalmology surgery scene has become an obstacle to the development of depth estimation in this scene. To resolve this problem, we built one synthetic dataset for cataract surgeries. The dataset contains information on RGB images, depth maps, and segmentation masks. We also adopt the state-of-the-art methods of depth estimation on this dataset as the baseline model to build the benchmark. We also analyze the generalization of the baseline models trained on the synthetic dataset to the real surgical scene.