Skip to content

Reference Implementations

Reference implementations are published datasets that show VIDS applied to real imaging data. Each one is archived on Zenodo with a DOI, so you can download it, run the validator yourself, and compare your result with the published one.

Under MD-0003, accepted in July 2026, a VIDS reference implementation publishes the VIDS metadata layer and its documentation; source images stay with their original repository and are cited by DOI. LIDC-Hybrid-100 was published before that policy and includes the CT volumes.

LIDC-Hybrid-100 (lung CT)

  • What it is: 100 lung CT volumes from LIDC-IDRI with consensus nodule segmentations derived from four-radiologist annotations, structured as a VIDS dataset
  • Annotation type: Segmentation
  • Validator result: 21 of 21 rules PASS under the Full profile, validator 1.2.1
  • Licence: CC BY 4.0
  • DOI: 10.5281/zenodo.19582717

Check it yourself:

pip install vids-validator
vids-validate lidc-hybrid-100 --profile full

The Validation Report page shows the full JSON output for this dataset.

VIDS-Fundus 1.0 (fundus photography, IDRiD)

  • What it is: The VIDS metadata layer for the Indian Diabetic Retinopathy Image Dataset (IDRiD): 443 fundus annotation sidecars covering grading and lesion segmentation
  • Annotation types: Classification and segmentation
  • Validator result: 15 of 21 rules PASS under validator 1.2.1
  • Source images: Not redistributed. Cited by the IDRiD DOI 10.21227/H25W98
  • Licence: CC BY 4.0
  • Reference implementation DOI: 10.5281/zenodo.21511086
  • Fundus specification DOI: 10.5281/zenodo.21488816

The fundus specification record defines the metadata model and annotation schema for fundus photography datasets and is the companion to the reference implementation above.

What a reference implementation shows

A validator result shows that a dataset satisfies the machine-checkable requirements of the selected profile, within the checks the validator implements. VIDS verifies that documentation is present and structured; it does not certify a dataset. It does not judge the clinical accuracy of any annotation.