{"success":true,"data":{"author":"Simula Research Laboratory and Baerum Hospital, Norway (Borgli et al.)","bucketAccessType":"public","bucketName":"SimulaMet-HOST/HyperKvasir","bucketPrefix":"","category":"Medicine","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:11.855Z","description":"HyperKvasir is a gastrointestinal endoscopy dataset, described by its publishers as the largest public one, collected from routine gastroscopy and colonoscopy examinations at Baerum Hospital in Norway and published by Simula Research Laboratory in Scientific Data (2020). The publisher's card lists 10,662 labelled images in 23 anatomical and pathological classes (landmarks such as the pylorus and cecum, findings such as polyps, esophagitis and ulcerative colitis, and therapeutic interventions), 99,417 unlabelled images, 1,000 polyp images with pixel masks and bounding boxes, and 374 endoscopy videos totalling 11.62 hours. Annotations were made by experienced gastroenterologists. As measured 2026-11-14 this repository, uploaded by Simula itself, holds four zip archives: labelled images (about 3.9 GB), segmented polyp images (46 MB), unlabelled images (about 29.4 GB) and videos (about 30.1 GB). Unzip to get JPEG images, JSON bounding boxes and AVI videos, then load them with PIL, OpenCV or torchvision. Used for polyp detection and segmentation, GI finding classification, anomaly detection and self-supervised learning on endoscopy video. Simula releases HyperKvasir under CC BY 4.0 as stated in the dataset card body (the YAML tag says MIT); cite Borgli et al. 2020.","displayTitle":"HyperKvasir: gastrointestinal endoscopy images and videos","id":"da9b5b4b-c4ab-56d5-8c09-c5540d46482a","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/SimulaMet-HOST/HyperKvasir","objectCountEstimate":6,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1038/s41597-020-00622-y","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"SimulaMet-HOST/HyperKvasir\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"SimulaMet-HOST/HyperKvasir\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=SimulaMet-HOST, license:mit, downloads=1389"],"format":"mixed","generatedAt":"2026-11-14T14:15:15.994Z","layout":{"example":"hyper-kvasir-labeled-images.zip","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"hyper-kvasir-labeled-images.zip","truncatedAt":512}],"sampleObjects":["hyper-kvasir-labeled-images.zip","hyper-kvasir-segmented-images.zip","hyper-kvasir-unlabeled-images.zip","hyper-kvasir-videos.zip"],"sidecars":[{"bytes":"---\nlicense: mit\n---\n# Dataset Card for HyperKvasir\n\n<!-- Provide a quick summary of the dataset. -->\n\nHyperKvasir is the largest publicly available dataset for gastrointestinal (GI) endoscopy, consisting of over 110,000 images and 374 videos, including labeled, unlabeled, and segmented samples. It supports tasks such as classification, segmentation, object detection, and anomaly detection in GI disease diagnostics.\n\n## Dataset Details\n\n### Dataset Description\n\nHyperKvasir is a comprehensive multi-class dataset collected from real gastro- and colonoscopy examinations at Bærum Hospital, Norway. It contains:\n- 10,662 labeled images across 23 classes\n- 99,417 unlabeled images\n- 1,000 polyp images with segmentation masks and bounding boxes\n- 374 endoscopy videos (11.62 hours)\n\n- **Curated by:** Simula Research Laboratory and Bærum Hospital, Norway\n- **Funded by [optional]:** Not specified\n- **Shared by [optional]:** Simula Research Laboratory\n- **Language(s) (NLP):** English (annotations and metadata)\n- **License:** Creative Commons Attribution 4.0 International (CC BY 4.0)\n\n### Dataset Sources [optional]\n\n- **Repository:** https://github.com/simula/hyper-kvasir\n- **Paper [optional]:** https://doi.org/10.1038/s41597-020-00622-y\n- **Demo [optional]:** https://datasets.simula.no/hyper-kvasir/\n\n## Uses\n\n### Direct Use\n\n- Development and benchmarking of computer vision models in medical imaging\n- Training models for GI tract disease detection\n- Medical education and training support\n- Polyp detection and segmentation research\n\n### Out-of-Scope Use\n\n- Direct clinical diagnosis or patient treatment decisions without expert review\n- Non-medical imaging tasks or commercial uses without adherence to license terms\n\n## Dataset Structure\n\n- **Labeled Images:** 10,662 images in 23 diagnostic and anatomical classes\n- **Unlabeled Images:** 99,417 unannotated GI endoscopy images\n- **Segmented Images:** 1,000 polyp images with corresponding masks (JPEG) and bounding boxes (JSON)\n- **Videos:** 374 AVI format videos with frame-level annotations in 171 sequences\n\n## Dataset Creation\n\n### Curation Rationale\n\nCreated to fill the lack of large, diverse, and annotated GI endoscopy datasets for AI research, especially for deep learning-based detection and classification tasks.\n\n### Source Data\n\n#### Data Collection and Processing\n\nCollected from routine clinical examinations using standard endoscopy equipment. Data were de-identified and curated, with videos annotated by expert gastroenterologists.\n\n#### Who are the source data producers?\n\nMedical professionals from Bærum Hospital, Norway, including gastroenterologists and technical staff.\n\n### Annotations [optional]\n\n#### Annotation process\n\nAnnotations were performed manually by trained gastroenterologists following clinical diagnosis protocols. Segmentations include bounding boxes and pixel-level masks for polyps.\n\n#### Who are the annotators?\n\nExperienced medical doctors and clinical experts in gastroenterology.\n\n#### Personal and Sensitive Information\n\nThe dataset has been anonymized and de-identified to protect patient privacy. No personally identifiable information is included.\n\n## Bias, Risks, and Limitations\n\n- Collected from a single hospital, so geographical and device variation is limited.\n- Some categories may be underrepresented, leading to class imbalance.\n- Videos are annotated for only a subset of findings.\n\n### Recommendations\n\nUsers should consider class imbalance and lack of multi-center data when generalizing model results. Medical models trained on this dataset should be validated on diverse clinical data.\n\n## Citation [optional]\n\n**BibTeX:**\n```\n@article{borgli2020hyperkvasir,\n  title={HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy},\n  author={Borgli, Hanna and Thambawita, Vajira and Smedsrud, Pia H and et al.},\n  journal={Scientific Data},\n  volume={7},\n  number={1},\n  pages={283},\n  year={2020},\n  publisher={Nature Publishing Group},\n  doi={10.1038/s41597-020-00622-y}\n}\n```\n\n**APA:**\n\nBorgli, H., Thambawita, V., Smedsrud, P","encoding":"utf8","key":"README.md","name":"README.md","size":4788,"truncated":true}],"sourceObjectKey":"hyper-kvasir-labeled-images.zip","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"SimulaMet-HOST/HyperKvasir\", filename=\"hyper-kvasir-labeled-images.zip\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"hyper-kvasir-labeled-images.zip","totalObjects":6},"summary":"HyperKvasir is a gastrointestinal endoscopy dataset, described by its publishers as the largest public one, collected from routine gastroscopy and colonoscopy examinations at…","title":"HyperKvasir: gastrointestinal endoscopy images and videos","updatedAt":"2026-11-14T17:28:12.095Z","visibility":"public","websiteUrl":"https://datasets.simula.no/hyper-kvasir/"}}