SankaraEyeHospital/Catacomp-104

CataCompDetect

Video dataset of cataract surgeries labeled for surgical complications (Posterior Capsule Rupture, Vitreous Loss, Iris Prolapse), used to train and evaluate the CataCompDetect complication-detection pipeline.

Dataset structure

catacomp-hf/
├── train/
│   ├── metadata.csv
│   ├── catacomp-train-001.mp4
│   ├── catacomp-train-002.mp4
│   └── ...
└── val/
    ├── metadata.csv
    ├── catacomp-val-001.mp4
    ├── catacomp-val-002.mp4
    └── ...

Videos are anonymized and renamed to public IDs of the form catacomp-<split>-<NNN> (1-based, zero-padded to 3 digits, assigned in original CSV order). No original identifiers or filenames are retained in this public release.

Splits

Split	Videos	Normal	Anomaly
train	53	32	21
val	51	32	19

Columns

Column	Description
video_id	Public video identifier (catacomp-<split>-NNN)
file_name	Relative path to the video file (used by datasets video loading)
label	Normal or Anomaly
pcr	Boolean flag: Posterior Capsule Rupture present
vitreous-loss	Boolean flag: Vitreous Loss present
iris-prolapse	Boolean flag: Iris Prolapse present
cortical_wash_start	Timestamp (mm:ss) — start of cortical wash stage
cortical_wash_stop	Timestamp (mm:ss) — end of cortical wash stage
iol_insertion_start	Timestamp (mm:ss) — start of IOL insertion stage

Loading

from datasets import load_dataset

ds = load_dataset("videofolder", data_dir="catacomp-hf")

Source

Generated from the original complication-data collection using the data/id_mapping.json bidirectional ID map in the CataCompDetect repository.