M. Alotaibi, M. Prabhushankar, K. Kokilepersaud, and G. AlRegib
Seismic interpretation is time-consuming and laborintensive. Machine learning (ML) offers a promising solution through its generalization capabilities from labeled data. However, ML methods demand vast amounts of data, which is expensive and time-consuming to obtain in the context of seismic interpretation. We hypothesize that cooperating ML methods with the domain expertise can address ML limitations. In particular, interactive prompting-based models like ChatGPT and the Segment Anything Model (SAM) enable this interactivity between AI and experts. To verify our hypotheses, we utilize SAM as part of our workflow. SAM is a segmentation model that segments any object through user-defined inclusion and exclusion points. This workflow, as illustrated in Fig. 1, allows the expert to utilize inclusion and exclusion points to segment the salt dome region. As they add or remove points, the system shows the resultant segmentation. Subsequently, they can target the regions that have been missed by SAM by adjusting these points without closing the prompting window.
