 This study compared two versions of a state-of-the-art object detection model, YOLOV5N and YOLOV5S, for their ability to detect cassava plants in a real-world setting. The models were tested on an embedded GPU system, allowing for real-time performance and accurate crop counting. The results showed that YOLOV5S outperformed YOLOV5N in terms of detection accuracy but was slower than YOLOV5N. YOLOV5S also performed better under weed interference, but at the cost of lower speed. This suggests that the choice of model should be made based on the desired trade-offs between detection speed, detection accuracy, and memory usage.