license: cc-by-4.0
Dataset Card for HyperKvasir
Dataset Summary
HyperKvasir is the largest publicly available gastrointestinal (GI) endoscopy dataset, comprising 110,079 images and 374 videos collected during real gastro- and colonoscopy examinations at Bærum Hospital in Norway. The dataset includes anatomical landmarks, pathological findings, and normal findings, making it suitable for various computer vision tasks in medical imaging.
Supported Tasks and Leaderboards
- ^Image Classification
- ^Object Detection
- ^Semantic Segmentation
- ^Anomaly Detection
- ^Video Summarization oai_citation:1‡datasets.simula.no oai_citation:2‡Papers with Code oai_citation:3‡Dataset Ninja
Languages
Dataset Structure
Data Splits
^The dataset is divided into the following subsets:
- Labeled Images: ^10,662 images across 23 classes (e.g., polyps, esophagitis, ulcers).
- Unlabeled Images: ^99,417 images without annotations.
- Segmented Images: ^1,000 polyp images with corresponding segmentation masks and bounding boxes.
- Annotated Videos: ^373 videos totaling approximately 11.62 hours and 1,059,519 frames, with 171 annotated findings. oai_citation:5‡datasets.simula.no
Data Format
- Images: ^JPEG format.
- Videos: ^AVI format.
- Segmentations: ^JPEG masks and JSON files for bounding boxes.
- Annotations: ^Provided in accompanying metadata files. oai_citation:6‡Academia oai_citation:7‡datasets.simula.no oai_citation:8‡datasets.simula.no
Dataset Creation
Curation Rationale
^The dataset was created to facilitate research in computer-aided diagnosis and detection in GI endoscopy by providing a large, diverse, and annotated dataset. oai_citation:9‡Academia
Source Data
- Collection: ^Data were collected during routine examinations using standard endoscopic equipment.
- Annotations: ^Performed by experienced gastroenterologists. oai_citation:10‡PMC
Annotations
- Labeled Images: ^Categorized into 23 classes based on findings.
- Segmented Images: ^Polyp regions manually segmented.
- Videos: ^Annotated for specific findings and landmarks. oai_citation:11‡datasets.simula.no
Licensing Information
Citation Information
^If you use this dataset, please cite:
^Borgli, H., Thambawita, V., Smedsrud, P.H., Hicks, S., Jha, D., Eskeland, S.L., Randel, K.R., Pogorelov, K., Lux, M., Nguyen, D.T.D., Johansen, D., Griwodz, C., Stensland, H.K., Garcia-Ceja, E., Schmidt, P.T., Hammer, H.L., Riegler, M.A., Halvorsen, P., & de Lange, T. (2020). HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy. Scientific Data, 7(1), 283. https://doi.org/10.1038/s41597-020-00622-y oai_citation:12‡simula.no
Contributions
- Data Collection: ^Bærum Hospital, Norway.
- Annotation: ^Experienced gastroenterologists.
- Dataset Preparation: ^Simula Research Laboratory. oai_citation:13‡ResearchGate
Dataset Links
- Homepage: ^https://datasets.simula.no/hyper-kvasir/
- Download: ^https://datasets.simula.no/downloads/hyper-kvasir/
- GitHub Repository: ^https://github.com/simula/hyper-kvasir oai_citation:14‡datasets.simula.no oai_citation:15‡OSF