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

Researchers have introduced a framework for zero-shot adaptation of the Segment Anything Model (SAM) for seismic interpretation, enhancing segmentation accuracy and reducing reliance on labeled data.

The Segment Anything Model (SAM) offers powerful zero-shot segmentation capabilities through prompt-based interaction^[1]. However, most existing applications of SAM rely on fine-tuning for specific geological targets, which requires extensive labeled data and incurs high computational cost. A new framework addresses this limitation by combining sparse user-defined point prompts with dense mask prompts derived from SAM's internal feature activations^[1]. This approach, along with geologic target-aware selection of seismic attributes and colormaps, improves boundary delineation and segmentation accuracy. The framework eliminates the need to retrain SAM for each geologic feature, reducing reliance on labeled data while preserving model generality. Another study on arxiv.org^[2] reported that SAM has demonstrated strong generalizability in various instance segmentation tasks, but its performance is dependent on the quality of manual prompts. The new framework, termed SPDA-SAM, has been shown to outperform state-of-the-art counterparts across twelve different data sets^[2].