TY - GEN
T1 - Benchmark
T2 - International Conference on Information Technology and Systems, ICITS 2025
AU - Jiménez-Delgado, E.
AU - Méndez-Porras, A.
AU - Campaña Bastidas, Sixto Enrique
AU - Calvo-Valverde, L.
AU - Alfaro-Velasco, J.
AU - Rodriguez-Salas, J.
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
PY - 2025
Y1 - 2025
N2 - Facial detection has become essential in fields like security and surveillance, requiring efficient and accurate systems. This study evaluates and compares six frameworks—RetinaFace, Mediapipe, YuNet, YoloFace, Haarcascade, and MTCNN—using the WIDER FACE dataset of 3,226 images. Performance is assessed on an Intel Core i5 13th generation 13600K CPU with 32 GB RAM, focusing on faces recognized, processing time, and average detection time per face. RetinaFace demonstrated the highest detection capability with 22,738 faces but required significant processing time (23,153.78 s, 3.83 s/face). In contrast, YuNet was the most efficient, averaging 0.03 s per face, followed by Mediapipe at 0.04 s, although with fewer detections. MTCNN and YoloFace showed intermediate performance in detection and efficiency. This study highlights key differences in deep learning-based facial detection frameworks, offering insights for future research and applications in the field.
AB - Facial detection has become essential in fields like security and surveillance, requiring efficient and accurate systems. This study evaluates and compares six frameworks—RetinaFace, Mediapipe, YuNet, YoloFace, Haarcascade, and MTCNN—using the WIDER FACE dataset of 3,226 images. Performance is assessed on an Intel Core i5 13th generation 13600K CPU with 32 GB RAM, focusing on faces recognized, processing time, and average detection time per face. RetinaFace demonstrated the highest detection capability with 22,738 faces but required significant processing time (23,153.78 s, 3.83 s/face). In contrast, YuNet was the most efficient, averaging 0.03 s per face, followed by Mediapipe at 0.04 s, although with fewer detections. MTCNN and YoloFace showed intermediate performance in detection and efficiency. This study highlights key differences in deep learning-based facial detection frameworks, offering insights for future research and applications in the field.
KW - AI
KW - Artificial intelligence
KW - Deep Learning
KW - Haarcascade
KW - MTCNN
KW - Mediapipe
KW - Neural network
KW - RetinaFace
KW - Wider Face
KW - YoloFace
KW - YuNet
UR - https://www.scopus.com/pages/publications/105012920534
U2 - 10.1007/978-3-031-93103-1_11
DO - 10.1007/978-3-031-93103-1_11
M3 - Contribución a la conferencia
AN - SCOPUS:105012920534
SN - 9783031931024
T3 - Lecture Notes in Networks and Systems
SP - 104
EP - 113
BT - Information Technology and Systems - ICITS 2025
A2 - Rocha, Alvaro
A2 - Ferrás, Carlos
A2 - Calvo, Hiram
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 22 January 2025 through 25 January 2025
ER -