Szczegóły publikacji
Opis bibliograficzny
Virtual HDR-based enhancement of cytological images for improved YOLO training / Maciej KUBEŁ, Bogusław CYGANEK // W: Computer Information Systems and Industrial Management : 25th international conference, CISIM 2026 : Bialystok, Poland, September 25–27, 2026 : proceedings / eds. Khalid Saeed, Jiří Dvorský. — Cham : Springer Nature, cop. 2027. — ( Lecture Notes in Computer Science ; ISSN 0302-9743 ; 16843 ). — ISBN: 978-3-032-38938-1; e-ISBN: 978-3-032-38939-8. — S. 3–14. — Bibliogr., Abstr. — Publikacja dostępna online od: 2026-09-22
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 170330 |
|---|---|
| Data dodania do BaDAP | 2026-10-02 |
| DOI | 10.1007/978-3-032-38939-8_1 |
| Rok publikacji | 2027 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Konferencja | International Conference on Computer Information Systems and Industrial Management Applications 2026 |
| Czasopismo/seria | Lecture Notes in Computer Science |
Abstract
The variability in illumination, staining, and dynamic range remains a major obstacle in training object detection models for cytological image analysis. This work investigates the use of Virtual High Dynamic Range (VHDR) enhancement as a preprocessing strategy to improve the quality and consistency of cytological images used for YOLO based detection. We introduce an automated optimization pipeline in which Optuna systematically searches the VHDR parameter space, using YOLO26s performance as the objective function. For each candidate configuration, dataset is enhanced, a YOLO26s model is retrained, and detection metrics are evaluated to guide the search towards high performance enhancement settings. Experimental results demonstrate that optimized VHDR preprocessing consistently improves precision, recall, and mAP scores compared to the baseline dataset. The findings confirm that performance driven enhancement optimization can meaningfully strengthen downstream detection accuracy and highlight VHDR as an effective preprocessing method for cytological imaging workflows.
