The article examines the advancements in computer vision through image segmentation, focusing on the fine-tuning of the YOLOv8 model for task-specific applications like distinguishing ducks in images. YOLOv8, part of the YOLO model family, is noted for its real-time object detection capabilities and significant improvements over its predecessor, YOLOv5. Fine-tuning involves adapting pre-trained models to new data sets, enhancing performance without the need for comprehensive retraining, thus saving time and computational resources. The process is demonstrated using the open images dataset and the Comet platform, which aids in tracking, logging, and storing experiments, highlighting the importance of fine-tuning in developing precise models for niche tasks. The experiment shows improvements over the pre-trained model, emphasizing the role of Comet in providing a comprehensive view of model performance and facilitating collaboration through detailed logging of metrics, configurations, and results.