Arianna De Vecchi, PhD candidate in Information Technology at the Department of Electronics, Information and Bioengineering – Politecnico di Milano and member of the Smart Eyewear Lab, has been selected as one of the three winners of the Best Paper Award at the workshop Eyes Of The Future: Integrating Artificial Intelligence in Smart Eyewear (IAISE), held within the international conference IJCNN 2025 – International Joint Conference on Neural Networks, which took place in Rome from June 30 to July 5, 2025.

Her paper, titled “Maximizing Eye Segmentation Accuracy with YOLO”, introduces an innovative approach to semantic segmentation of eye images, leveraging YOLO models—typically used for object detection—to accurately identify the three main regions of the eye: sclera, iris, and pupil. The work presents a semi-automatic labeling pipeline, based on a large-scale YOLO model, to efficiently generate segmentation masks from a small initial set of manual labels. These masks are then used to train lightweight (nano) versions of YOLOv8 and YOLOv11.

The resulting models demonstrate strong generalization capabilities on images captured with hardware different from the one used during training, as well as under more challenging visual conditions, such as partially closed eyes or infrared images.

The research was carried out in collaboration with EssilorLuxottica within the Smart Eyewear Lab, a Joint Research Centre between EssilorLuxottica and Politecnico di Milano. The study also benefited from the contributions of researcher Sabrina Azzi, PhD candidate Pietro Bartoli, researcher Marco Paracchini, and Prof. Federica Villa.