Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts

The study introduces a framework for zero-shot adaptation of the Segment Anything Model (SAM) for seismic interpretation, which enhances segmentation accuracy without the need for extensive fine-tuning. Key components include aligning seismic attributes and visualization techniques with geological targets, and a hybrid prompting strategy that combines user-defined point prompts with dense mask prompts from SAM's internal feature activations. This approach allows practitioners to utilize SAM effectively across various geological contexts while minimizing the dependency on labeled data, thus streamlining the application of large pretrained models in seismic analysis.