Self-adaptive deployment of synthetic-trained object detectors via test-time adaptation
Yazarlar (5)
Dr. Öğr. Üyesi Zehra ELMI Beykoz Üniversitesi, Türkiye
Furkan Ali Turan
İstanbul Sabahattin Zaim University, Türkiye
İstanbul Sabahattin Zaim University, Türkiye
Emir Tuna
İstanbul Sabahattin Zaim University, Türkiye
İstanbul Sabahattin Zaim University, Türkiye
Abdullatif Akkaya
İstanbul Sabahattin Zaim University, Türkiye
İstanbul Sabahattin Zaim University, Türkiye
Soheila Elmi
Koç University, Türkiye
Koç University, Türkiye
| Makale Türü |
Özgün Makale
(SSCI, AHCI, SCI, SCI-Exp dergilerinde yayınlanan tam makale)
|
||
| Dergi Adı | Machine Vision and Applications (Q3) | ||
| Dergi ISSN | 0932-8092 Dergi Bilgileri (2026) | ||
| Dergi Tarandığı Indeksler | SCI-Expanded | ||
| Makale Dili | Türkçe | Basım Tarihi | 06-2026 |
| Kabul Tarihi | 24-05-2026 | Yayınlanma Tarihi | 06-06-2026 |
| Cilt / Sayı / Sayfa | 37 / 4 / – | DOI | 10.1007/s00138-026-01849-2 |
| Makale Linki | https://link.springer.com/article/10.1007/s00138-026-01849-2 | ||
| UAK Araştırma Alanları |
Görüntü İşleme
Yapay Zeka
|
||
| Özet |
| Deploying synthetic-trained object detectors in real environments is challenging under single-image, batch= 1 inference, where pseudo-label noise and unstable normalization can undermine online adaptation. We present a deployment-oriented test-time adaptation framework for synthetic-to-real object detection that requires neither source data nor target labels. The framework combines entropy-guided adaptation triggering, IoU-filtered pseudo-labeling with augmentation consistency and greedy matching, and architecture-aware normalization updates for both CNN-and Transformer-based detectors. Trained on LookSIM synthetic data and evaluated on 3,500 real defense images spanning six target classes, the method improves from 51.7 to 56.5 for RF-DETR and from 49.7 to 54.9 for YOLOv12-L while preserving real-time deployment on RTX 3090 and Jetson AGX Orin. Ablation results show that IoU … |
| Anahtar Kelimeler |
| Domain adaptation | Edge deployment | Real-time object detection | Synthetic data | Test-time adaptation |
BM Sürdürülebilir Kalkınma Amaçları
| Atıf Sayıları | |
| Google Scholar | 1 |