2026-10-15T09:32:00Z https://data.mendeley.com/oai

oai:data.mendeley.com/ts94gkj79z.1 2026-09-15T10:59:43Z

Sinha, Ujjwal Gastrointestinal-Disease-Detection Mendeley Data Total Images: 1,196 endoscopic images Datasets: 2 combined sources Format: JPEG images with corresponding segmentation masks 1. Kvasir-SEG Dataset Images: 1,000 endoscopic images Structure: images/ - Original endoscopic images masks/ - Pixel-level segmentation masks (1,000 masks) kavsir_bboxes.json - Bounding box annotations in JSON format Features: Variable resolutions (typically 500-1400px width, 400-1100px height) Multiple polyps per image possible Bounding box coordinates: [xmin, ymin, xmax, ymax] Labels: "polyp" annotations File sizes: ~25KB to ~200KB per image Annotation Format: { "image_id": { "height": 529, "width": 622, "bbox": [ {"label": "polyp", "xmin": 38, "ymin": 5, "xmax": 430, "ymax": 338} ] }} 2. Kvasir-Sessile Dataset Images: 196 endoscopic images Structure: images/ - Original endoscopic images masks/ - Segmentation masks (196 masks) Focus: Sessile (flat) polyp detection Purpose: Complementary training data for enhanced polyp detection Dataset Details CSV File The dataset_details.csv contains metadata for all 1,196 images: Columns: dataset - Source dataset (kvasir-seg or kvasir-sessile) image_filename - Image file name image_path - Full path to image mask_path - Full path to corresponding mask image_width - Image width in pixels image_height - Image height in pixels image_size_bytes - Image file size mask_size_bytes - Mask file size has_mask - Whether mask exists (Yes/No) num_bounding_boxes - Number of bounding boxes per image bounding_boxes - Bounding box coordinates (for kvasir-seg) label - Polyp label bbox_width / bbox_height - Bounding box dimensions Dataset Characteristics Image Resolutions: Common sizes: 528×622, 530×622, 546×626, 531×570 pixels High-resolution examples: 1070×1348, 1011×1220 pixels Standardized to 512×512 for training (as per project documentation) Polyp Distribution: Most images contain 1 polyp Some images contain multiple polyps (2-11 bounding boxes) All images are labeled as "polyp" (positive cases) Annotation Types: Segmentation Masks - Pixel-level polyp regions (binary masks) Bounding Boxes - Rectangular coordinates (kvasir-seg only) Labels - Polyp classification Usage in Project Training Split: 70% (837 images) Validation Split: 15% (179 images) Test Split: 15% (180 images) Classes: Binary classification (Polyp vs. No Polyp) Applications: Deep learning model training (YOLO11m, CNN classifiers) Polyp detection and segmentation Clinical research in gastroenterology Dataset Purpose Designed for automated gastrointestinal polyp detection in endoscopic images, supporting: Computer vision research Medical AI development Clinical decision support systems Polyp detection and segmentation model training The dataset provides both pixel-level segmentation masks and bounding box annotations, suitable for both segmentation and detection tasks Gastrointestinal Disorder Gastrointestinal Cancer Gastrointestinal Complication Gastrointestinal Bleeding Gastrointestinal Nursing Gastrointestinal Pathology Gastrointestinal Finding Swain, Kaliprasanna Dataset doi:10.17632/ts94gkj79z.1 oai:data.mendeley.com/ts94gkj79z.1 Creative Commons Attribution 4.0 International http://creativecommons.org/licenses/by/4.0 https://data.mendeley.com/datasets/ts94gkj79z 2026-09-15T10:59:43Z