Deep convolution neural network with scene-centric and object-centric information for object detection

摘要

In recent years, Deep Convolutional Neural Network (CNN) has shown an impressive performance on computer vision field. The ability of learning feature representations from large training dataset makes deep CNN outperform traditional hand-crafted features approaches on object classification and detection. However, computations for deep CNN models are time consuming due to their high complexity, which makes it hardly applicable to real world application, such as Advance Driver Assistance System (ADAS). To reduce the computation complexity, several fast object detection frameworks in the literature have been proposed, such as SSD and YOLO. Although these kinds of method can run at real-time, they usually struggle with dealing of small objects due to the difficulty of handling smaller input image size. Based on our observation, we propose a novel object detection framework which combines the feature representations learned from object-centric and scene-centric datasets with an aim to improve the accuracy on detecting especially small objects. The experimental results on MSCOCO dataset show that our method can actually improve the detection accuracy of small objects, which leads to better overall results. We also evaluate our method on PASCAL VOC 2012 datasets, and the results show that our method not only can achieve state-of-the-art accuracy but also most importantly presents in real-time.