Abstract—The rapid advancement of technology has significantly enhanced object detection tasks that require high speed and accuracy. Although the YOLOv8n model is known for its efficient and accurate detection capabilities, its detection process still involves multiple complex stages, limiting its deployment in challenging environments. This research investigates how the YOLOv8n architecture can be modified to enhance inference speed without compromising detection accuracy. The objective is to propose a modification to the YOLOv8n backbone by integrating a gating mechanism on top of the C2f module. The model, YOLOv8nGM, utilizes a sigmoid activation function to implement gating, converting network outputs into probabilities that enable selective computation depending on input complexity. This adaptive mechanism enables the network to bypass unnecessary calculations, thereby reducing computational overhead. Experiments on the benchmark dataset reveal that YOLOv8nGM achieves an average inference time of 8.72 ms per image, representing an approximately 5.4% improvement in speed compared to the original YOLOv8n. Additionally, the GPU memory consumption was reduced from 0.27 GB to 0.21 GB, signifying more efficient resource use. The mAP50 metrics on the validation and test datasets showed that the proposed model achieved minor improvements of 0.02% and 0.008%, respectively, compared to the original model, indicating that accuracy was maintained despite faster computation. These findings demonstrate that the gating mechanism effectively balances inference speed and precision, making the YOLOv8nGM model suitable for real-time object detection applications such as warehouse inventory counting. This study advances object detection technology by providing a practical and computationally efficient solution for environments that require rapid and accurate recognition.
 
Keywords—deep learning, YOLOv8n, gating mechanism, object detection


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).