Integrated Multi-Source Dataset and Reproducibility Code for Colorectal Polyp Detection with YOLOv11 Annotations and LOF Preprocessing

This repository contains processed versions of five publicly available colorectal polyp datasets: CVC-ColonDB, CVC-ClinicDB, Kvasir-SEG, ETIS, and EndoScene (CVC-300). The original datasets were publicly accessible. In this repository, segmentation masks were converted into bounding box annotations suitable for training YOLOv11-based object detection models.The data were further cleaned using a Local Outlier Factor (LOF) preprocessing method to remove anomalous samples. This processed dataset is intended to support reproducible experiments in colorectal polyp detection. The dataset includes images, masks, and labels folders for each original dataset. This version also includes the Python implementation of the LOF-YOLOv11n framework used in the associated manuscript. The code package contains scripts for deterministic five-fold dataset partitioning, LOF-based training-data filtering, YOLOv11n training and evaluation, per-dataset performance analysis, and supplementary characterization of LOF-flagged images. Installation and usage instructions are provided in the included README file.