|Keywords||small object detector, super-resolved generative adversarial network, object detection, object recognition, small objectm region of interest|
|Publication||Master Graduation thesis in College of Engineering Seoul National University 2020|
The small object detector refers to an object detector that is specialized to detect small objects in an image. These small object detectors are becoming an important research area since they are used in various industries such as self-driving, fashion, military, security and health care.
The reason why object detectors show relatively low performance for smaller objects than different size objects is due to a size of input image and the limits of expression of the extracted image characteristics.
To solve this problem, various methods have recently been studied and shown good performance, however, these have caused problems such as the need for a lot of computational resources and the inability to end-to-end learning.
This study proposes small object detection based on generative adversarial neural network to overcome the problem that needs many computational resources or multiple-step learning processes while maintaining good performance.
Finally, detecting medium and small object was improved while maintaining overall performance 𝑚𝐴𝑃 in similar level.