Cataract-LMM: Large-Scale Multi-Source Multi-Task Cataract Surgery Benchmark

3,000 cataract surgery procedures, 1,134.2 hours of video with phase annotations, instance segmentation, tracking, and skill scoring. Largest public cataract-surgery-video resource.

At a glance

Field	Value
Short name	cataract_lmm
Full name	Cataract-LMM: Large-Scale Multi-Source Multi-Task Cataract Surgery Benchmark
Primary category	surgical_video
Contained modalities	surgical_video
Tasks	classification, segmentation, multilabel
Samples	3,000
Classes	Not reported (Not reported)
Splits	train, val, test
Size	200.0 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	anonymous_direct
Route backend	HuggingFace Hub
Availability	available (checked 2026-07-21)
Acquisition support	standard_platform_supported
Legacy sample-loader status	Metadata and access only

Notes

Largest public cataract-surgery-video resource by hours. CC BY-NC-ND 4.0 — non-commercial + no derivatives on HF.

Access preflight and acquisition

CLI

# Read-only preflight
eyehub download cataract_lmm --data-dir ./data --dry-run --json

# Explicit transfer, only when preflight reports supported behavior
eyehub download cataract_lmm --data-dir ./data

Python

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

ds = REGISTRY.get_dataset('cataract_lmm')
print(preflight_dataset(ds, './data'))  # no transfer

Upstream page: huggingface.co/datasets

Source-term evidence: huggingface.co/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{cataract_lmm,
  title  = { Cataract-LMM: Large-Scale Multi-Source Multi-Task Cataract Surgery Benchmark },
  note   = { Ahmadi MJ, Gandomi I, Abdi P, et al., 'Cataract-LMM: Large-scale multi-source multi-task benchmark for deep learning in surgical video analysis', Scientific Data 2026. doi:10.1038/s41597-026-07464-0 },
  year   = { 2026 },
  url    = { https://huggingface.co/datasets/mjahmadi/Cataract-LMM },
}

Plain text

Ahmadi MJ, Gandomi I, Abdi P, et al., 'Cataract-LMM: Large-scale multi-source multi-task benchmark for deep learning in surgical video analysis', Scientific Data 2026. doi:10.1038/s41597-026-07464-0

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

• ophora: Ophora-160K: Ophthalmic Surgical Video Instruction Dataset (160,185 records, unknown)
• lmod_plus: LMOD+ Multimodal Ophthalmology Benchmark (32,633 records, unknown)
• cadis: CaDIS: Cataract Dataset for Image Segmentation (4,670 records, research-only)
• ophnet2024: OphNet2024: Ophthalmic Surgical Video Dataset (2,278 records, cc-by-nc-sa)
• cataract1k: Cataract-1K: Large-Scale Cataract Surgery Video Dataset (1,000 records, research-only)
• migs_video: Multicenter Fine-Annotated MIGS Surgical Video Dataset (186 records, cc-by)
• sics155: SICS-155 Small Incision Cataract Surgery Videos (155 records, cc-by-nc)
• cataract101_extended_labels: Cataract-101 Extended Labels (101 records, cc-by)