<![CDATA[Abstrak − Ekspresi wajah merupakan indicator penting dalam interaksi manusia yang dapat dianalisis secara otomatis menggunakan teknologi deteksi objek. Penelitian ini bertujuan membandingkan performa YOLOv5, YOLO11, dan YOLO12 dalam mendeteksi tujuh kelas ekspresi wajah pada 2.935 dataset public. Model dilatih dengan konfigurasi yang seragam, kemudian di evaluasi berdasarkan nilai mean Average Precision (mAP) dan latensi inferensi. YOLOv5 mencatat nilai mAP tertinggi pada saat proses training dan validation, sedangkan YOLO11 memiliki lantensi terendah. Pada evaluasi model, YOLO12 unggul dalam nilai mAP, dan YOLO11 tetap tercepat dalam latensi. Model di deployment dengan 6 data citra yang memiliki 7 kelas. Hasil menunjukkan bahwa meskipun model YOLO12 unggul dalam akurasi evaluasi model, YOLO11 lebih optimal dari segi kecepatan inferensi.Kata Kunci: face expressions; accuracy-latency; yolo12; Abstract − Facial expression is an important indicator of human interaction that can be analyzed automatically using object detection technology. This study aims to compare the performance of YOLOv5, YOLO11, and YOLO12 in detecting seven classes of facial expressions on 2,935 public datasets. The models were trained with a uniform configuration, and then evaluated based on the mean Average Precision (mAP) value and inference latency. YOLOv5 recorded the highest mAP value during training and validation, while YOLO11 had the lowest latency. On model evaluation, YOLO12 excelled in mAP value, and YOLO11 remained the fastest in latency. The model was deployed with 6 image data that had 7 classes. Results show that while the YOLO12 model excels in model evaluation accuracy, YOLO11 is more optimal in terms of inference speed.Keywords: face expressions; accuracy-latency; yolo12]]>
Copyrights © 2025
