First sent to members by email on October 20, 2026. Added to this website when it launched on October 26, 2026.
Ten months after the Plate Archive Project started, it has its first results.
Over that time members labeled 14,600 small cutouts taken from 312 DASCH plates. Each cutout was marked as a star, one of several kinds of defect, or a handwritten annotation. We used that set to train a classifier, then ran it over about 41,000 detections from 1,200 plates.
The model agrees with our human reviewers 93 percent of the time. Its most common mistake is confusing very faint stars with small grains in the emulsion, which is also the case our reviewers disagreed about most often.
Members can now use:
- the labeled cutouts and their labels,
- the classifier and its weights, and
- the catalog of classified detections.
We would like to publish all three. Before we do, a second group of members will review a random sample of the catalog, and we want to make sure the release follows the DASCH data terms.
The method and results are described in ASRU-TN-2026-003. None of it would exist without the work of the DASCH project, which is grateful for partial support from NSF grants AST-0407380, AST-0909073 and AST-1313370.
Thank you to every member who spent an evening labeling.