iMED: A Multi-Endoscope Dataset for Surgical 3D Perception

DOI

iMED v1.0 is the first release of the iMED dataset. It contains one representative scene per session, enabling users to inspect the dataset structure, metadata, video format, calibration files, timestamps, tool masks when available, and reference trajectories. This release is not the full iMED dataset. The full dataset release is deferred while the MICCAI iMED challenge is active to reduce data leakage. Challenge submissions are due on Sept. 1, with the planned full iMED release targeted for the ECCV conference date, Sept. 8.The v1.0 subset is approximately 6 GB. The full dataset contains approximately seven scenes per session and be approximately 45 GB.Included data:processed stereo videosraw stereo videostimestamp CSV filesintrinsics and calibration filescalibration sequences for recomputation workflowsdataset, session, scene, and sequence metadatainstrument masks when tools are presentreference trajectories ArUco dictionary metadata Stereo depth maps are not included due to storage size. They are intended to be recomputable from the released videos, calibration files, and accompanying code located at: https://github.com/surgical-vision/imed

Identifier
DOI https://doi.org/10.5522/04/32764638.v1
Related Identifier HasPart https://ndownloader.figshare.com/files/66080882
Metadata Access https://api.figshare.com/v2/oai?verb=GetRecord&metadataPrefix=oai_datacite&identifier=oai:figshare.com:article/32764638
Provenance
Creator Bonilla, Sierra
Publisher University College London UCL
Contributor Figshare
Publication Year 2026
Rights https://creativecommons.org/licenses/by/4.0/; http://purl.org/coar/access_right/c_f1cf
OpenAccess true
Contact researchdatarepository(at)ucl.ac.uk
Representation
Language English
Resource Type Dataset
Discipline Other