Harvard-FairVision (AMD + DR + Glaucoma, paired SLO + OCT)

30,000 subjects (10K each AMD, DR, glaucoma) with paired SLO fundus and OCT B-scans, demographic attributes (race, ethnicity, gender, language), for fairness analysis.

At a glance

Field	Value
Short name	harvard_fairvision
Full name	Harvard-FairVision (AMD + DR + Glaucoma, paired SLO + OCT)
Primary category	multimodal
Contained modalities	fundus, oct
Tasks	classification
Samples	30,000
Classes	3 (amd, dr, glaucoma)
Splits	train, val, test
Size	600.0 GB
Source-stated terms	CC BY-NC-ND 4.0
Normalized terms	cc-by-nc-nd
Descriptive screening label	Explicit noncommercial clause recorded; check source
Terms scope	dataset_files
Access friction	controlled_or_manual
Route backend	Manual (upstream-gated)
Availability	available (checked 2026-07-21)
Acquisition support	manual_access_blocked
Legacy sample-loader status	Metadata and access only

Notes

Application-gated (Harvard form). No automated mirror. Sub-repos: github.com/Harvard-Ophthalmology-AI-Lab/Harvard-{AMD,DR,Glaucoma}.

Access preflight and acquisition

CLI

# This route requires upstream human action; no transfer starts.
eyehub download harvard_fairvision --data-dir ./data --dry-run --json
# Follow the official instructions shown by preflight.

Python

from eyedatahub.acquisition import preflight_dataset
from eyedatahub.datasets.registry import REGISTRY

ds = REGISTRY.get_dataset('harvard_fairvision')
print(preflight_dataset(ds, './data'))  # returns manual_access_blocked

Upstream page: ophai.hms.harvard.edu/datasets

Source-term evidence: ophai.hms.harvard.edu/datasets

Loader status

This catalog record provides metadata and access instructions, but it does not yet include a standard DatasetSample loader. Inspect the source file structure or contribute a loader before using it in a training pipeline.

Citation

BibTeX

@misc{harvard_fairvision,
  title  = { Harvard-FairVision (AMD + DR + Glaucoma, paired SLO + OCT) },
  note   = { Luo et al., 'FairVision: Equitable Deep Learning for Eye Disease Screening via Fair Identity Scaling', arXiv 2310.02492; Harvard Ophthalmology AI Lab 2024. Three disease sub-repos: Harvard-AMD, Harvard-DR, Harvard-Glaucoma. 30,000 subjects total (10K each) },
  year   = { 2024 },
  url    = { https://ophai.hms.harvard.edu/datasets/harvard-fairvision30k },
}

Plain text

Luo et al., 'FairVision: Equitable Deep Learning for Eye Disease Screening via Fair Identity Scaling', arXiv 2310.02492; Harvard Ophthalmology AI Lab 2024. Three disease sub-repos: Harvard-AMD, Harvard-DR, Harvard-Glaucoma. 30,000 subjects total (10K each).

Source-stated terms

• Raw source string: CC BY-NC-ND 4.0
• Normalized category: cc-by-nc-nd
• Apparent scope: dataset_files
• Descriptive screening label: Explicit noncommercial clause recorded; check source

⚠️ Source-stated terms, scope, and normalized labels are curation metadata, not legal advice or a permission finding. Review the current official source before transfer or reuse.

Related datasets with shared modalities

• eyecare_100k: Eyecare-100K: Multimodal Ophthalmology VQA Corpus (102,000 records, unknown)
• multieye: MultiEYE: OCT-Enhanced Fundus Multi-Disease Benchmark (58,036 records, mit)
• lmod_plus: LMOD+ Multimodal Ophthalmology Benchmark (32,633 records, unknown)
• mmrdr: MMRDR: Multi-Modal Retinal Diabetic Retinopathy Dataset (24,460 records, cc-by)
• x_pcr: X-PCR Ophthalmology Progressive Clinical Reasoning Benchmark (18,700 records, unknown)
• olives: OLIVES: Ophthalmic Labels for Investigating Visual Eye Semantics (9,408 records, cc-by)
• oct_fundus_dme_dr_mexico: OCT and Eye Fundus Dataset for DME and DR (2,661 records, unknown)
• grape: GRAPE: Glaucoma Real-world Appraisal Progression Ensemble (1,115 records, cc0)