BROWSE
Detail
Conference Paper
A Study on Military Object Detection in Panoramic View Using Stable Diffusion
Cited 0 time in 
Share 
Share
- Authors
- Jiwon Lee, Sungwon Moon, Dowon Nam
- Issue Date
- 2023-10
- Citation
- International Conference on Information and Communication Technology Convergence (ICTC) 2023, pp.1889-1891
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- The bottleneck in the development of military object detectors based on deep learning is the costly and high-difficulty training data collection. Although various methods have been proposed to solve this problem, augmentation methods for sparse data, which is extremely difficult to collect, is still an area of high difficulty that requires continuous research. This paper proposes a method to generate training data of an object detector for military smart glasses with a panoramic view with extremely insufficient learning data by using outpainting with stable diffusion. Through experiments, it has been proved that the proposed method has far superior training performance than training data generated by simply cropping or concatenating existing training data in the form of a panoramic view. The proposed method can be used as one of the excellent data augmentation methods in a situation where an image of unusual size, such as a fish-eye view image or an ultra-wide-angle image, is required as training data.
- KSP Keywords
- Data Augmentation, Data Collection, Fish-eye, Learning data, Object detection, Wide-angle, deep learning(DL), object detector, panoramic view, smart glasses, sparse data