Yale-BIDS-Chen/VOLMO-Evaluation-Data

VOLMO: Versatile and Open Large Models for Ophthalmology
Evaluation Data

---

📋 Overview

Evaluation data of VOLMO: Versatile and Open Large Models for Ophthalmology.

✨ Dataset Features

• 📸 30GB+ Images - Retinal fundus photos, OCT scans, and clinical imaging
• 🎯 4 Task Categories - Binary classification, staging, description, and clinical assessment
• 📝 24 Benchmark Datasets - Standardized evaluation across multiple conditions
• 🔬 Clinical Annotations - Expert-involved labels and descriptions

---

🚀 Download Data

The VOLMO evaluation datasets are hosted on Hugging Face:

Option 1: Using Hugging Face Datasets (Recommended)

from datasets import load_dataset

# Load entire dataset
dataset = load_dataset("Yale-BIDS-Chen/VOLMO-Evaluation-Data")

Option 2: Using Hugging Face Hub

from huggingface_hub import snapshot_download

# Download entire dataset
data_path = snapshot_download(
    repo_id="Yale-BIDS-Chen/VOLMO-Evaluation-Data",
    repo_type="dataset",
    local_dir="./data"
)

Option 3: Using Git LFS

# Install git-lfs if not already installed
git lfs install

# Clone the dataset repository
git clone https://huggingface.co/datasets/Yale-BIDS-Chen/VOLMO-Evaluation-Data

Option 4: Manual Download

Visit https://huggingface.co/datasets/Yale-BIDS-Chen/VOLMO-Evaluation-Data to download individual files.

---

📁 Data Structure

data/
├── datajsons/                     # Dataset annotations (JSON format)
│   ├── bool/                      # Binary classification (17 datasets)
│   │   ├── brset_amd.json
│   │   ├── brset_dr.json
│   │   ├── biobank_amd_tianyi.json
│   │   └── ...
│   ├── stage/                     # Disease staging (4 datasets)
│   │   ├── eyepacs.json
│   │   ├── mhstage.json
│   │   └── ...
│   ├── imgdesc/                   # Image descriptions (1 dataset)
│   │   └── pmc_serina.json
│   └── assessplan/                # Clinical assessments (1 dataset)
│       └── pmc_casereports.json
│
└── images/                        # Image files (30GB total)
    ├── brset/                     # BRSET dataset (4.2GB)
    ├── biobank/                   # UK Biobank (24GB)
    ├── eyepacs/                   # EyePACS DR (9.0GB)
    ├── sydney_innovation/         # Sydney Innovation (8.0GB)
    ├── fives/                     # FIVES glaucoma (117MB)
    ├── sustech/                   # SuSTech (469MB)
    ├── oimhs/                     # OIMHS OCT (185MB)
    ├── pmc_casereports/           # PMC case reports (31MB)
    └── pmc_eyes/                  # PMC-Eyes (6.1MB)

---

🎯 Task Categories

🔵 Binary Classification (17 Datasets)

Detection of presence/absence for various retinal conditions:

Dataset	Samples	Condition
BRSET AMD	108	Age-Related Macular Degeneration
BRSET DR	442	Diabetic Retinopathy
BRSET Hypertension	102	Hypertensive Retinopathy
BRSET ICDR	1,322	Increased Cup-Disc Ratio
BRSET Nevus	56	Choroidal Nevus
BRSET Macular Edema	142	Macular Edema
UK Biobank AMD	400	AMD Screening
UK Biobank DR	241	DR Screening
UK Biobank Glaucoma	400	Glaucoma Screening
...	...	...

Total: ~5,000+ samples across 17 conditions

📊 Stage Classification (4 Datasets)

Multi-class disease severity grading:

Dataset	Samples	Task
EyePACS	8,871	DR Severity (0-4)
MH Stage	900	Macular Hole Staging
Sydney Innovation	7,982	DR Classification
SuSTech SYSU	1,151	DR Grading

Total: ~19,000 samples

📝 Image Description (1 Dataset)

Natural language descriptions of clinical findings:

• PMC Serina: 41 retinal images with detailed clinical descriptions

🏥 Assessment & Plan (1 Dataset)

Clinical decision support annotations:

• PMC Case Reports: 89 cases with assessment and treatment plans

---

📊 Data Format

JSON Annotation Format

Each JSON file contains a list of samples with the following structure:

{
  "image_paths": ["path/to/image.jpg"],
  "prompt": "Does this image show signs of diabetic retinopathy?",
  "GT": "Yes",
}

---

📈 Dataset Statistics

Category	Datasets	Total Samples	Total Size
Binary Classification	17	~5,000	~28GB
Stage Classification	4	~19,000	~17GB
Image Description	1	41	~6MB
Assessment & Plan	1	89	~31MB
Total	24	~24,000+	~30GB

---

📝 License

This dataset is licensed according to the original licenses of data.

---

🙏 Acknowledgments

This dataset compilation includes data from multiple sources:

• BRSET
• OIMHS
• UK Biobank
• EyePACS
• Sydney Innovation
• FIVES
• PMC
• etc...

We thank all original data providers and contributors.

---

📧 Contact

For questions or collaboration inquiries, please email:

• Zhenyue Qin: kf.zy.qin@gmail.com
• Qingyu Chen: qingyu.chen@yale.edu

---

🔗 Resources

• 🏠 Main Repository: https://github.com/kfzyqin/volmo
• 🤗 Model Weights: https://huggingface.co/Yale-BIDS-Chen/VOLMO-2B
• 📊 This Dataset: https://huggingface.co/datasets/Yale-BIDS-Chen/VOLMO-Evaluation-Data

---

Built with ❤️ for exploring AI's potential in ophthalmology

VOLMO: Versatile and Open Large Models for Ophthalmology