Test data for ICPR 2026 - RARE-Vision Competition

Le Floch, Maxime

Test data for ICPR 2026 - RARE-Vision Competition

Type of the data datacite.resourceTypeGeneral	Dataset
Total size of the dataset datacite.size	52666435203
Author dc.contributor.author	Le Floch, Maxime
Upload date dc.date.accessioned	2026-03-09T18:11:50Z
Publication date dc.date.available	2026-03-09T18:11:50Z
Publication date dc.date.issued	2026-03-09
Abstract of the dataset dc.description.abstract	The RARE-VISION test dataset consists of three previously unseen capsule endoscopy examinations acquired with the Navicam system at the University Hospital Carl Gustav Carus, Technical University of Dresden. The dataset is strictly separated from the development data, and ground-truth annotations are withheld for final evaluation. Each case is provided as a complete chronological video sequence at a resolution of 480 × 480 pixels, preserving the original temporal order without trimming or manual segmentation. The three videos contain: 44,878 frames 53,220 frames 62,927 frames The data reflect real-world clinical variability and the natural class imbalance of capsule endoscopy, where rare pathological findings occur sparsely within long sequences of normal mucosa. Annotations are defined as temporal events (start frame, end frame, label) corresponding to the 17 competition target classes. The label set includes anatomical regions and pathological findings only; no anatomical landmarks are annotated. The dataset is designed to evaluate robust rare-event detection, temporal consistency, and fully automatic inference on long sequential video streams. The videos are provided exclusively for scientific research within the scope of the ICPR 2026 RARE-VISION competition and must not be used for any commercial purposes. For detailed terms of use, please refer to the official competition report and documentation. This study was approved by the Ethics Committee of the University Hospital Carl Gustav Carus at the Technical University of Dresden on December 16, 2022 (Ethics ID: BO-EK-534122022), confirming adherence to the ethical principles of the Declaration of Helsinki. Due to the retrospective anonymization of the data and their collection during clinically indicated routine interventions, explicit consent was not required. This is additionally supported by the Ethics Committee’s approval, a consultation with the data privacy officer, and local law. Section 34, Paragraph 1 of the Saxon Hospital Act (SächsKHG) explicitly allows the collection and analysis of this type of data.
Public reference to this page dc.identifier.uri	https://opara.zih.tu-dresden.de/handle/123456789/2119
Public reference to this page dc.identifier.uri	https://doi.org/10.25532/OPARA-1119
Publisher dc.publisher	Technische Universität Dresden
Licence dc.rights	Attribution-NonCommercial-NoDerivatives 4.0 International	en
URI of the licence text dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/4.0/
Specification of the discipline(s) dc.subject.classification	2::22::205::205-15
Title of the dataset dc.title	Test data for ICPR 2026 - RARE-Vision Competition
Project abstract opara.project.description	Video Capsule Endoscopy (VCE) is widely used for the non-invasive examination of the gastrointestinal tract. However, a single examination may produce tens of thousands of frames, making manual review by clinicians time-consuming and error-prone. Automated analysis using machine learning has the potential to support clinical workflows, yet it remains challenging due to severe class imbalance, visual variability, and the rarity of clinically relevant findings. The RARE-VISION Challenge focuses on advancing automated analysis of VCE data with particular emphasis on the detection and classification of rare gastrointestinal findings. Participants are invited to develop algorithms capable of identifying anatomical regions and pathological patterns from capsule endoscopy imagery. The challenge dataset consists of annotated VCE frames collected from real clinical examinations and includes a wide range of normal anatomical structures as well as rare abnormalities. The competition aims to benchmark state-of-the-art machine learning methods under realistic clinical conditions characterized by noisy imagery and highly imbalanced class distributions. Submitted methods will be evaluated using standardized metrics designed to reflect both overall performance and sensitivity to rare classes. By providing a common dataset and evaluation framework, the RARE-VISION Challenge seeks to stimulate research on robust and clinically relevant approaches for automated gastrointestinal video analysis.
Funding Acknowledgement opara.project.fundingAcknowledgement	Funding Semeco A4 This research was funded by the BMBF (German Federal Ministry of Education and Research) as part of the SEMECO cluster4future FKZ 03ZU1210GA and 03ZU121HB.
Public project website(s) opara.project.publicReference	https://github.com/RAREChallenge2026/RARE-VISION-2026-Challenge
Project title opara.project.title	ICPR 2026 - RARE-Vision Competition