<h3>目标检测经典论文翻译汇总：[<a href="https://blog.csdn.net/Jwenxue/article/details/107737752">翻译汇总</a>]</h3>
<h3><a name="t1"></a>翻译pdf文件下载：<a href="https://github.com/bigcindy/DeepLearningPapersTranslation">[下载地址]</a></h3>
<h3><a name="t2"></a>此版为纯中文版，中英文对照版请稳步：[<a href="https://blog.csdn.net/Jwenxue/article/details/107846360">R-FCN纯中文版</a>]</h3>
<table border="1" cellspacing="0"><tbody><tr><td colspan="4" style="text-align:center;vertical-align:middle;width:103.7pt;"> <h1 style="margin-left:0cm;"><strong><span style="color:#000000;">R-FCN: Object Detection via Region-based Fully Convolutional Networks</span></strong></h1> <h1 style="margin-left:0cm;"><strong><span style="color:#000000;">基于</span></strong><strong><span style="color:#000000;">Region</span></strong><strong><span style="color:#000000;">的全卷积神经网络进行目标检测</span></strong></h1> </td></tr><tr><td style="vertical-align:top;width:103.7pt;"> <p style="margin-left:0cm;"><span style="color:#000000;">Jifeng Dai</span></p> <p style="margin-left:0cm;"><span style="color:#000000;">Microsoft Research</span></p> </td><td style="vertical-align:top;width:103.7pt;"> <p style="margin-left:0cm;"><span style="color:#000000;">Yi Li*</span></p> <p style="margin-left:0cm;"><span style="color:#000000;">Tsinghua University</span></p> </td><td style="vertical-align:top;width:103.7pt;"> <p style="margin-left:0cm;"><span style="color:#000000;">Kaiming He（何恺明）</span></p> <p style="margin-left:0cm;"><span style="color:#000000;">Microsoft Research</span></p> </td><td style="vertical-align:top;width:103.7pt;"> <p style="margin-left:0cm;"><span style="color:#000000;">Jian Sun（</span><span style="color:#000000;">孙剑）</span></p> <p style="margin-left:0cm;"><span style="color:#000000;">Microsoft Research</span></p> </td></tr></tbody></table>
<h2 style="margin-left:0cm;"><span style="color:#ff0000;">Abstract</span></h2>
<p style="text-indent:33px;"><span style="color:#000000;">We present region-based, fully convolutional networks for accurate and efficient object detection. In contrast to previous region-based detectors such as Fast/Faster R-CNN [6, 18] that apply a costly per-region subnetwork hundreds of times, our region-based detector is fully convolutional with almost all computation shared on the entire image. To achieve this goal, we propose position-sensitive score maps to address a dilemma between translation-invariance in image classification and translation-variance in object detection. Our method can thus naturally adopt fully convolutional image classifier backbones, such as the latest Residual Networks (ResNets) [9], for object detection. We show competitive results on the PASCAL VOC datasets (e.g., </span><span style="color:#000000;">83.6% </span><span style="color:#000000;">mAP on the 2007 set) with the 101-layer ResNet. Meanwhile, our result is achieved at a test-time speed of 170ms per image, 2.5-20</span><span style="color:#000000;">×</span><span style="color:#000000;"> faster than the Faster R-CNN counterpart. Code is made publicly available at: </span><a href="https://github.com/daijifeng001/r-fcn"><span style="color:#000000;">https://github.com/daijifeng001/r-fcn</span></a><span style="color:#000000;">.</span></p>
<h2 style="margin-left:0cm;"><span style="color:#ff0000;">摘要</span></h2>
<p style="text-indent:33px;"><span style="color:#000000;">我们提出了基于</span><span style="color:#000000;">Region</span><span style="color:#000000;">的全卷积网络，以实现准确和高效的目标检测。与先前的基于区域的检测器（如</span><span style="color:#000000;">Fast/Faster R-CNN [6</span><span style="color:#000000;">，</span><span style="color:#000000;">18]</span><span style="color:#000000;">）相比，这些检测器每次需要对每个</span><span style="color:#000000;">region</span><span style="color:#000000;">运行非常耗时子网络，我们的基于区域的检测器是全卷积的，几乎所有计算都在整张图像上共享。为了实现这一目标，我们提出了<strong>位置敏感分数图（</strong></span><span style="color:#000000;">position-sensitive score maps</span><strong><span style="color:#000000;">）</span></strong><span style="color:#000000;">，以解决图像分类中的平移不变性与目标检测中的平移可变性之间的矛盾。因此，我们的方法自然地采用全卷积图像分类器的主干网络，如最新的残差网络（</span><span style="color:#000000;">ResNets</span><span style="color:#000000;">）</span><span style="color:#000000;">[9]</span><span style="color:#000000;">，用于目标检测。我们使用</span><span style="color:#000000;">101</span><span style="color:#000000;">层</span><span style="color:#000000;">ResNet</span><span style="color:#000000;">在</span><span style="color:#000000;">PASCAL VOC</span><span style="color:#000000;">数据集上获得了具有竞争力的结果（例如，</span><span style="color:#000000;">2007</span><span style="color:#000000;">数据集上</span><span style="color:#000000;">83.6%</span><span style="color:#000000;">的 |
|
转播
