| Title |
Analysis Study of Performance and Limitations of Each Image Matching Deep Learning Network using Mobile Backbone |
| Authors |
박인호(Inho Park) ; 이현재(Hyunjae Lee) ; 안우현(Woohyun Ahn) ; 김국병(Kukbyung Kim) ; 김영광(Young-gwang Kim) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.8.25 |
| Keywords |
Image matching; Mobile device; Mobile backbone; CNN |
| Abstract |
The trend of image matching has made continuous progress from techniques using traditional feature extraction to techniques using deep learning. Among them, the deep learning technique can derive various performances depending on the shape of the network, and has a significant advantage in image matching when an appropriate backbone and loss function are selected. In this paper, considering the problem that CNN-based image matching networks use backbone networks that are heavy to apply to mobile devices, various mobile backbones are applied to matching networks. In addition, through these mobile backbones and existing deep learning networks, we compare and analyze the performance between the networks to which each mobile backbone is applied and the existing networks, and analyze the limitations of CNN-based deep learning matching networks at mobile device. A matching network with a mobile backbone is applied to a mobile device through TFLite to compare and analyze the performance. Through these attempts, this paper will help researchers to easily access and analyze image matching networks for mobile devices that require high speed and low capacity. |