Within the broader remit of the International Rock Imaging Summit, this workshop supports ongoing efforts to strengthen coordination and shared practice across the rock imaging community. While the meeting spans multiple modalities and length scales—from the millimetre to the atomic—this workshop focuses on X‑ray computed tomography (XCT), which has become a foundational tool in rock imaging. Despite its widespread adoption for exploring the three‑dimensional structure of rocks, our community still lacks shared standards, interoperable training datasets, and open, reusable machine‑learning models for segmentation. This interactive workshop aims to bring together researchers, practitioners, and software developers to collaboratively define what a community‑driven ecosystem for XCT segmentation should look like—and to take the first concrete steps toward building it. The workshop will be distributed across the four days of the meeting, allowing participants to respond to conference talks, refine ideas, test model components, and contribute to the structure of a community white paper.

The workshop will begin by establishing a roadmap for best practices in sharing, curating, and documenting XCT datasets for machine‑learning applications. Participants will work together to identify essential metadata, evaluate existing standards (including insights from established domains such as bioimaging), and define a practical checklist for dataset preparation and exchange. Through group discussion, digital polling, and a post‑it‑style ideation board, we will also identify priority training and educational needs—ranging from imaging fundamentals to tutorials, workshops, and method‑focused publications.

Building on this foundation, the second part of the workshop focuses on the technical pathway toward a community‑developed, open, and shareable neural network for XCT segmentation. Participants will explore how voxel scale, lithological diversity, and material classes (e.g., sandstone vs. shale, gabbro vs. basalt, porous media subclasses) influence model generalization and reusability. We will collaboratively outline strategies for generating and sharing labels, assembling representative training datasets, and designing a base model architecture that can be broadly applied and easily fine‑tuned.

By the end of the conference, attendees will have co‑authored the framework for a white paper on community standards for XCT machine learning and contributed to the blueprint for the first open, extensible neural network for XCT segmentation. This workshop is designed for anyone interested in imaging, machine learning, data standards, or building sustainable open‑science tools within the XCT community—and will also lay the groundwork for expanding these efforts to additional imaging modalities in future meetings.