We propose an object detection method that improves the accuracy of theconventional SSD (Single Shot Multibox Detector), which is one of the topobject detection algorithms in both aspects of accuracy and speed. Theperformance of a deep network is known to be improved as the number of featuremaps increases. However, it is difficult to improve the performance by simplyraising the number of feature maps. In this paper, we propose and analyze howto use feature maps effectively to improve the performance of the conventionalSSD. The enhanced performance was obtained by changing the structure close tothe classifier network, rather than growing layers close to the input data,e.g., by replacing VGGNet with ResNet. The proposed network is suitable forsharing the weights in the classifier networks, by which property, the trainingcan be faster with better generalization power. For the Pascal VOC 2007 testset trained with VOC 2007 and VOC 2012 training sets, the proposed network withthe input size of 300 x 300 achieved 78.5% mAP (mean average precision) at thespeed of 35.0 FPS (frame per second), while the network with a 512 x 512 sizedinput achieved 80.8% mAP at 16.6 FPS using Nvidia Titan X GPU. The proposednetwork shows state-of-the-art mAP, which is better than those of theconventional SSD, YOLO, Faster-RCNN and RFCN. Also, it is faster thanFaster-RCNN and RFCN.
展开▼
机译：我们提出了一种对象检测方法，可提高Chononventional SSD（单拍摄Multibox检测器）的精度，这是一种精度和速度方面的TopObject检测算法之一。众所周知，随着特色数量的增加，已知深网络的表现。但是，很难通过简单地提高特征图的数量来提高性能。在本文中，我们提出并分析了如何有效地使用特征地图，以提高常规系统的性能。通过更改结构关闭托盘型网络，而不是靠近输入数据的层，例如，通过更换与Reset的vggnet来获得增强的性能。所提出的网络适用于分类器网络中的权重，其中属性，TrainingCan具有更好的泛化功率。对于用VOC 2007和VOC 2012训练训练的Pascal VOC 2007测试集，所提出的网络，输入大小为300 x 300，在35.0 fps（每秒帧）时，实现了78.5％的地图（平均平均精度），而网络使用NVIDIA Titan X GPU，512 x 512尺寸达到80.8％的地图，以16.6 fps地图。 ProposedNetwork显示出最先进的地图，它比Theconventional，Yolo，Faster-RCNN和RFCN更好。此外，它是速度速度超过rcnn和rfcn。
展开▼
