COCO 2017 object detections with YOLOv5 confidence scores: a probabilistic image-retrieval database

A probabilistic image-retrieval database: the objects detected by a YOLOv5 model on the images of the COCO 2017 dataset (training and validation sets), together with the detector's confidence scores, plus the COCO ground-truth annotations of the same images. It is meant as a benchmark for probabilistic databases and provenance management, and ships with scripts loading it into PostgreSQL and into ProvSQL. The data was built for the EDBT 2025 demonstration “Using a Probabilistic Database in an Image Retrieval Application” and is the dataset used in the experiments of “Provenance of HAVING Queries in Semirings with Monus”. The tables are identical to those distributed in the demonstration's source repository; this record adds documentation, a stable identifier and loading scripts. See README.md for the schema, the per-split counts, how the stained images were produced, and loading instructions.