Shanghai Eye Disease Center Ophthalmic Multimodal Dataset (上海市眼病防治中心多模态眼科数据集)

A comprehensive, multi-center, longitudinal ophthalmic foundation dataset from Shanghai Eye Disease Prevention & Treatment Center.

Repository Layout

Kaphathy/Shanghai/
└── Topcon/
    └── shards/
        ├── manifest.json              # O(1) Index mapping each exam_id to its shard
        ├── meta.tar.gz                # Complete clinical JSON metadata for all 30,711 exams (3.3 MB)
        ├── labels/                    # 62 TAR shards containing 2,133,834 pixel-level segmentation masks (26 GB)
        │   ├── labels_shard_000.tar
        │   └── ... labels_shard_061.tar
        └── images/                    # 62 TAR shards containing 2,195,256 B-scans, Fundus & SLO (727 GB)
            ├── images_shard_000.tar
            └── ... images_shard_061.tar

Dataset Statistics (Topcon DRI-OCT Triton Cohort)

• Total Valid Exams: 30,711 examinations
• Structural OCT B-scans: 2,133,834 slices (1050 nm Swept-Source OCT)
• True Color Fundus Photography (CFP): 30,711 high-resolution images
• Infrared SLO / Alignment Localizer: 30,711 images
• Pixel-level 9-Layer Retinal Segmentation Masks: 2,133,834 PNG masks
• Longitudinal Follow-up: 7 continuous years (2018–2024), 1,141 verified longitudinal patient chains

Quick Start (Loading via Python / WebDataset)

import json, tarfile
from huggingface_hub import hf_hub_download

# 1. Download index manifest
manifest_path = hf_hub_download(repo_id="Kaphathy/Shanghai", repo_type="dataset", filename="Topcon/shards/manifest.json")
with open(manifest_path, "r", encoding="utf-8") as f:
    manifest = json.load(f)

# 2. Check which shard contains exam 10001
exam_id = "10001"
shard_name = manifest["exams"][exam_id] # e.g. shard_020

# 3. Download corresponding image shard
tar_path = hf_hub_download(
    repo_id="Kaphathy/Shanghai",
    repo_type="dataset",
    filename=f"Topcon/shards/images/images_{shard_name}.tar"
)

# 4. Stream or extract directly
with tarfile.open(tar_path, "r") as tar:
    for member in tar.getmembers():
        if f"images/{exam_id}/" in member.name:
            f = tar.extractfile(member)
            # read image bytes...