Comparative Evaluation of ShuffleNetV2 and CSPResNeXt-50 Backbone Architectures for YOLO11-Based Face Detection
Küçük Resim Yok
Tarih
2025
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Institute of Electrical and Electronics Engineers Inc.
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
In today's hyper-connected world of technology, there is a huge flow of data. Video data from security cameras in surveillance systems constitutes a significant portion of this data stream. Fast and accurate face detection for variopus resolutions in image data continues to be crucial in a wide range of fields, particularly in autonomous vehicles and security camera systems. Face detection studies have been conducted on various You Only Look Once (YOLO) architectures in the literature. This study redesigned the backbone and header structures based on YOLO11 model, aiming to improve face detection performance in terms of accuracy, speed, and model complexity. To this end, four different scales of ShuffleNetV2 and CSPResNeXt-50 backbones are integrated into the YOLO11 model. FPN-like headers are configured with ShuffleNetV2, and PANet-like headers are configured with CSPResNeXt. A YOLO-compatible version of the Wider Face dataset was used in the study. As a result of the comparisons, the YOLO11 + CSPResNeXt-50 architecture demonstrated the highest performance with 90% precision, 87% sensitivity, and 97.1% mAP@50. © 2025 IEEE.
Açıklama
10th International Conference on Computer Science and Engineering, UBMK 2025 -- 17 September 2025 through 21 September 2025 -- Istanbul -- 214243
Anahtar Kelimeler
backbone, face detection, yolo11
Kaynak
International Conference on Computer Science and Engineering, UBMK
WoS Q Değeri
Scopus Q Değeri
N/A
Cilt
Sayı
2025
