Gastrointestinal-Disease-Detection
Description
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