REG2026 (REG²) is a Grand Challenge competition for pathology report generation from gigapixel whole slide images (WSIs), with diagnostic reasoning made explicit—not only the final report. It extends REG2025 by evaluating structured reasoning aligned with how pathologists explore slides and decide what is reportable.
The release pairs WSIs (TIFF, 20×) with chain-of-thought Q&A and CAP-style report fields (~12K training cases across seven organs). Phases, downloads, rules, and scoring are on the challenge overview and data description.
| Directory | What it is |
|---|---|
algorithm_submission_template/ |
Docker container template for submitting your algorithm to Grand Challenge |
submission_evaluation_code/ |
Local Docker evaluation harness — run the official scoring logic on your own machine before submitting |
- Open
algorithm_submission_template/. - Follow
algorithm_submission_template/README.md(Docker, both task interfaces, local tests, upload). - For weights layout, see
algorithm_submission_template/model/README.md.
Use submission_evaluation_code/ to run the same metric A + metric B scoring that the leaderboard uses — entirely on your own machine, with no upload required.
- Open
submission_evaluation_code/. - Follow
submission_evaluation_code/README.mdfor the full setup (Docker image, judge model weights, data formats, and three-step pipeline). - Drop in your own cases and predictions to get a realistic score estimate before the official submission deadline.