YOLO26 Object Detection Model — Reproducible Training, Evaluation & Results Repository

This repository contains the complete experimental package used to train, evaluate, and analyze a YOLO26 object detection model for solar feature recognition. It is organized into modular components that ensure full reproducibility of the training and evaluation pipeline. The Model directory contains the final trained model checkpoint (best weights), representing the optimal configuration obtained after hyperparameter tuning. The Ray_tune directory includes the JSON files required to reproduce the hyperparameter optimization process using Ray Tune, along with the corresponding analysis of the search results and the Jupyter notebook used for this stage. The Train_model notebook provides the complete training pipeline using the optimal hyperparameter configuration, while the Test_model notebook is used to evaluate the final model and generate performance metrics. The SDO directory contains real prediction outputs on observational solar data, enabling qualitative assessment of model performance under realistic conditions. Together, these components allow full reconstruction of the experimental workflow, from hyperparameter optimization to final evaluation on a held-out test set. The repository is intended for research, benchmarking, educational purposes, and further development of computer vision methods applied to solar physics.