The Kvasir dataset is part of the Medical Multimedia Challenge proposed by MediaEval and is mainly used for research on computer-aided digestive disease detection. The dataset contains 8,000 endoscopic images, divided into 8 categories, each containing 1,000 image examples. These images are mainly obtained from the gastrointestinal tract based on endoscopic procedures and are annotated and verified by professional doctors.

The 8 categories are based on the polypectomy procedure and are:

• 3 landmarks: z-line, pylorus, cecum
• 3 pathological findings: esophagitis, polyps, ulcerative colitis
• 2 other categories: dyed and lifted polyps, dyed resection margins