| Title |
Model Ensemble for Speed Enhancement in Object Detection |
| Authors |
이진수(Jin-su Lee) ; 이상광(Sang-Kwang Lee) ; 양성일(Seong-Il Yang) |
| DOI |
https://doi.org/10.5573/ieie.2019.56.6.35 |
| Keywords |
Object detection ; Ensemble method ; Convolutional neural network |
| Abstract |
Object detection is a research field about locating and classifying objects in visual scenes by analysing the visual information. It has been applied in a variety of industrial fields such as autonomous driving and image surveillance. At the early stage of research, a method that compares the features in given image and designed features had been used. Recently, object detection methods based on machine learning, especially convolutional neural network, has shown outstanding performance, compared to the traditional detection methods. In order to increase the accuracy in detection with deep learning method, multiple CNN models can be used jointly, which is called ensemble method, however, with the speed in detection decreasing. In this paper, we propose an ensemble method that increases the speed in object detection, compared to existing ensemble method. The increasement in speed is acquired as concatenating the feature maps which are the output of neural network models. It is proved with experiments that the speed in object detection increases with proposed ensemble method. More efficient way to control GPU memory is expected to enhance the performance in object detection with proposed ensemble method. |