Skripsi

OPTIMALISASI DETEKSI WAJAH KECIL BERBASIS YOLOV11 MENGGUNAKAN SAHI DAN REAL-ESRGAN

---

This study addresses the critical challenge of micro-scale face detection in lowresolution surveillance imagery by proposing a hybrid pipeline integrating YOLOv11-pose, Slicing Aided Hyper Inference (SAHI), and Real-ESRGAN using a two-phase methodology on the WIDER FACE dataset enriched with 5-point landmark annotations. The first phase focuses on architectural optimization, where YOLOv11s-pose was selected as the best model due to its optimal balance between parameter count and accuracy. The second phase, conducted through an ablation study, demonstrates that the SAHI strategy significantly improves detection accuracy (mean Average Precision) in the Hard category by 15.5% and on smalldegraded faces by 23.1%, validating the effectiveness of patch-based processing. Conversely, integrating Real-ESRGAN as a pre-processing step proved ineffective for machine detection due to domain mismatch, despite drastically enhancing perceptual visual quality with a 22.8% improvement in the BRISQUE score. In conclusion, the study recommends the YOLOv11s + SAHI configuration for automated detection optimization, while Real-ESRGAN is allocated as a postprocessing method for human visual verification

---

Availability

Detail Information

Series Title

-

Call Number

T1899522025

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2025

Collation

xv, 136 hlm.; ilus.; tab.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.370 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Teknik Informatika
Deteksi Wajah--Yolov11

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment