CIG-Bench: A Comprehensive Survey and Benchmark for AI-Driven Subsurface Imaging Understanding

Subsurface imaging understanding bridges observed geophysical data and quantitative geological models, supporting hydrocarbon exploration, CO2 storage site assessment, and geohazard monitoring. Over the past decade, deep learning has substantially reshaped interpretation workflows. To take stock of this progress, we systematically analyze 652 publications from 2015 to 2025 and organize the field into four major tasks: structural interpretation, geobody identification, seismic facies analysis, and property estimation. Yet subsurface imaging interpretation differs fundamentally from other AI-driven tasks, facing ambiguous signals, pronounced interpretive non-uniqueness, sparse semantics, unfixed target locations, and scarce reliable annotations. Building on the reviewed literature, we summarize three interrelated challenges that define its frontier: interpretation under complex geological conditions, cross-survey semantic generalization under low information density, and the absence of reliable benchmarks. Addressing them will hinge on integrating human expertise, physical constraints, and geological priors into model training or inference, and on treating uncertainty quantification as an intrinsic model output. Among these, the lack of unified benchmarks has been particularly consequential, making fair comparison difficult, hindering reproducibility, and fragmenting community efforts. We therefore propose a community benchmark spanning fault segmentation, relative geologic time estimation, geobody segmentation, and property modeling. It integrates unified evaluation protocols, pretrained models, and datasets that combine synthetic data for quantitative evaluation with real surveys for qualitative assessment. By coupling a decade-spanning review with an evolving benchmark, this work offers a timely reference and a reproducible foundation to accelerate future research and deployment.