Assessing the Impact of Layer Selection on CAM-Based Explainability for YOLOv8: A Study on Hand-Sketched Digital Logic Circuits

Abstract

Object detection has recently become a cornerstone task in computer vision. It allows the machine to identify and localise multiple objects in an image. With the development of deep learning, object detection systems have achieved remarkable performance in various domains such as autonomous driving, surveillance, healthcare and industrial automation. Among the various detection frameworks, the YOLO (You Only Look Once) family of models has attracted significant attention due to its unified architecture, real-time inference capabilities, and high detection accuracy \cite{redmon2016you}. YOLO models frame object detection as a single regression problem , directly from image pixels to bounding box coordinates and class probabilities . This makes them very fast and efficient for real world usage. The Artificial Intelligence evolution has quickly transitioned from Narrow Intelligence (ANI) to the present day of sophisticated deep learning and Transformers. As AI systems become increasingly autonomous and intelligent, the research community has shifted from pure accuracy to 'Explainable AI' (XAI) as pointed out in current trend reports \cite{singh2023introduction}. Our work is based on this global trend, by considering the YOLOv8 detector not as a simple output generator, but as a system to explain its detections to the user.