一种应用于目标检测中同时满足上述所有约束条件的全卷积神经网络结构。在我们的网络中，使用卷积层不只是用来提取特征图，同时也是作为输出层去计算边界框（bounding box）和分类概率。我们模型中的检测管道（detection pipeline）只包含一个神经网络的前向通路，因此它运行起来是极其迅速的。我们的模型是全卷积结构的，因此可以达到小的模型规模和很高的能量利用效率。最后的实验表明我们的模型能达到很高的精度，在 KITTI 基准上达到了最高的精确度。(A fully convoluted neural network structure applied to target detection simultaneously satisfying all the above constraints. In our network, using convolution layer is not only used to extract characteristic figure, but also as the output layer to compute bounding box (bounding box) and the classification probability. The detection pipeline in our model contains only a forward path of a neural network, so it runs extremely quickly. Our model is fully convoluted, so small model sizes and high energy efficiency can be achieved. The experimental results show that our model can achieve high precision, and achieve the highest accuracy on the KITTI benchmark.)