CRAC-DM: class relation-aware categorical diffusion model for surgical scene segmentation

Key Points

• The aim is to enhance segmentation accuracy and efficiency in surgical scene analysis.
• Developed a class relation-aware categorical diffusion model.
• Enhanced the forward process using inter-class similarity.
• Improved the reverse process with a deterministic S2D and targeted TTA.
• Achieved superior segmentation accuracy compared to existing models.
• Showed enhanced efficiency and reliability in surgical scene segmentation.

Abstract