Single Image Motion Deblurring for Mobile Devices

Kuspys, Nazarii

Home
→
Students Research & Project Works | Роботи студентів
→
Факультет прикладних наук
→
Бакалаврська програма "Комп'ютерні науки"
→
2023
→
View Item

dc.contributor.author	Kuspys, Nazarii
dc.date.accessioned	2024-02-14T11:46:22Z
dc.date.available	2024-02-14T11:46:22Z
dc.date.issued	2023
dc.identifier.citation	Kuspys, Nazarii. Research of Data Augmentation Approaches for Enhancing Classification Model Performance / Nazarii Kuspys; Supervisor: Volodymyr Karpiv; Ukrainian Catholic University, Department of Computer Sciences. – Lviv: 2023. – 41 p.: ill.	uk
dc.identifier.uri	https://er.ucu.edu.ua/handle/1/4431
dc.language.iso	en	uk
dc.title	Single Image Motion Deblurring for Mobile Devices	uk
dc.type	Preprint	uk
dc.status	Публікується вперше	uk
dc.description.abstracten	Motion blur is a common issue in image processing and video production. It is cru- cial to have a profound pre-processing method to address this obstacle for various image processing applications, e.g., object detection. Many existing state-of-the-art methods are limited to their computational complexity and memory requirements if one wants to use them in practice on mobile devices. This work proposes sev- eral ideas to overcome this issue and optimize existing solutions via architectural changes and graph optimizations. Moreover, we introduced an adapted version of Soft Attention inside skip connections and achieved a PSNR raise of 0.22 (dB) for the selected baseline. Finally, we derived the optimized model for mobile real-time applications without a significant drop in accuracy, e.g., obtaining 31.36 dB PSNR on GoPro (for image deblurring) with 24 FPS on the mobile application.	uk