计算机视觉经典论文

  • k6_789099
    了解作者
  • 45MB
    文件大小
  • zip
    文件格式
  • 0
    收藏次数
  • VIP专享
    资源类型
  • 0
    下载次数
  • 2022-04-09 23:28
    上传日期
计算机视觉:alexnet ,vgg ,resnet ,rcnn ,faster-rcnn mask-rcnn paper
computer_vision_classic_paper.zip
  • computer_vision_classic_paper
  • MultiNet.pdf
    12MB
  • Faster R-CNN.pdf
    6.6MB
  • AlexNet.pdf
    1.3MB
  • R_CNN.pdf
    6.2MB
  • Res_Net.pdf
    800.2KB
  • Visualizing_and_Understanding.pdf
    2.2MB
  • SPP_NET.pdf
    2.2MB
  • SSD.pdf
    2.4MB
  • MaskRCNN.pdf
    7.3MB
  • VGG.pdf
    195.3KB
  • GoogleNet.pdf
    214.3KB
  • NIN.pdf
    581.4KB
  • YOLO.pdf
    5.1MB
  • Fast-RCNN.pdf
    725.4KB
  • FCN.pdf
    372.7KB
内容介绍
<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/6252280c74bc5c0105ba53e9/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6252280c74bc5c0105ba53e9/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">MultiNet:<span class="_ _0"> </span>Real-time<span class="_"> </span>Joint<span class="_"> </span>Semantic<span class="_"> </span>Reasoning<span class="_"> </span>f<span class="_ _1"></span>or<span class="_"> </span>A<span class="_ _1"></span>utonomous<span class="_"> </span>Dri<span class="_ _2"></span>ving</div><div class="t m0 x2 h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">Marvin<span class="_"> </span>T<span class="_ _3"></span>eichmann</div><div class="t m0 x3 h4 y3 ff2 fs2 fc0 sc0 ls0 ws0">1<span class="_ _4"></span>2<span class="_ _4"></span>3</div><div class="t m0 x4 h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">,<span class="_"> </span>Michael<span class="_"> </span>W<span class="_ _3"></span>eber</div><div class="t m0 x5 h4 y3 ff2 fs2 fc0 sc0 ls0 ws0">2</div><div class="t m0 x6 h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">,<span class="_"> </span>Marius<span class="_"> </span>Z</div><div class="t m0 x7 h3 y4 ff2 fs1 fc0 sc0 ls0 ws0">&#168;</div><div class="t m0 x8 h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">ollner</div><div class="t m0 x9 h4 y3 ff2 fs2 fc0 sc0 ls0 ws0">2</div><div class="t m0 xa h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">,<span class="_"> </span>Roberto<span class="_"> </span>Cipolla</div><div class="t m0 xb h4 y3 ff2 fs2 fc0 sc0 ls0 ws0">3</div><div class="t m0 xc h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">and<span class="_"> </span>Raquel<span class="_"> </span>Urtasun</div><div class="t m0 xd h4 y3 ff2 fs2 fc0 sc0 ls0 ws0">1</div><div class="t m0 x3 h4 y5 ff2 fs2 fc0 sc0 ls0 ws0">1</div><div class="t m0 xe h3 y6 ff2 fs1 fc0 sc0 ls0 ws0">Department<span class="_"> </span>of<span class="_"> </span>Computer<span class="_"> </span>Science,<span class="_"> </span>Uni<span class="_ _2"></span>versity<span class="_"> </span>of<span class="_"> </span>T<span class="_ _5"></span>oronto</div><div class="t m0 xf h4 y7 ff2 fs2 fc0 sc0 ls0 ws0">2</div><div class="t m0 x10 h3 y8 ff2 fs1 fc0 sc0 ls0 ws0">FZI<span class="_"> </span>Research<span class="_"> </span>Center<span class="_"> </span>for<span class="_"> </span>Information<span class="_"> </span>T<span class="_ _3"></span>echnology<span class="_ _3"></span>,<span class="_"> </span>Karlsruhe</div><div class="t m0 xe h4 y9 ff2 fs2 fc0 sc0 ls0 ws0">3</div><div class="t m0 x11 h3 ya ff2 fs1 fc0 sc0 ls0 ws0">Department<span class="_"> </span>of<span class="_"> </span>Engineering,<span class="_"> </span>Uni<span class="_ _2"></span>versity<span class="_"> </span>of<span class="_"> </span>Cambridge</div><div class="t m0 x12 h5 yb ff3 fs3 fc0 sc0 ls0 ws0">marvin.teichmann@googlemail.com,<span class="_ _6"> </span>Michael.Weber@fzi.de,</div><div class="t m0 x13 h5 yc ff3 fs3 fc0 sc0 ls0 ws0">zoellner@fzi.de,<span class="_ _6"> </span>rc10001@cam.ac.uk,<span class="_ _6"> </span>urtasun@cs.toronto.edu</div><div class="t m0 x14 h6 yd ff1 fs1 fc0 sc0 ls0 ws0">Abstract</div><div class="t m0 x15 h7 ye ff4 fs4 fc0 sc0 ls0 ws0">While<span class="_ _7"> </span>most<span class="_ _7"> </span>appr<span class="_ _1"></span>oaches<span class="_ _7"> </span>to<span class="_ _7"> </span>semantic<span class="_ _7"> </span>r<span class="_ _1"></span>easoning<span class="_ _7"> </span>have<span class="_ _7"> </span>fo-</div><div class="t m0 x16 h7 yf ff4 fs4 fc0 sc0 ls0 ws0">cused<span class="_ _8"> </span>on<span class="_ _8"> </span>impr<span class="_ _1"></span>oving<span class="_ _8"> </span>performance,<span class="_ _9"> </span>in<span class="_ _8"> </span>this<span class="_ _8"> </span>paper<span class="_ _8"> </span>we<span class="_ _8"> </span>ar<span class="_ _1"></span>gue</div><div class="t m0 x16 h7 y10 ff4 fs4 fc0 sc0 ls0 ws0">that<span class="_"> </span>computational<span class="_"> </span>times<span class="_"> </span>are<span class="_"> </span>very<span class="_"> </span>important<span class="_"> </span>in<span class="_"> </span>or<span class="_ _1"></span>der<span class="_"> </span>to<span class="_"> </span>en-</div><div class="t m0 x16 h7 y11 ff4 fs4 fc0 sc0 ls0 ws0">able<span class="_ _a"> </span>r<span class="_ _1"></span>eal<span class="_"> </span>time<span class="_ _a"> </span>appl<span class="_ _2"></span>ications<span class="_ _a"> </span>such<span class="_ _a"> </span>as<span class="_ _a"> </span>autonomous<span class="_ _a"> </span>driving<span class="_ _2"></span>.<span class="_ _b"> </span>T<span class="_ _5"></span>o-</div><div class="t m0 x16 h7 y12 ff4 fs4 fc0 sc0 ls0 ws0">war<span class="_ _1"></span>ds<span class="_ _8"> </span>this<span class="_ _8"> </span>goal,<span class="_ _8"> </span>we<span class="_ _8"> </span>pr<span class="_ _1"></span>esent<span class="_ _8"> </span>an<span class="_ _8"> </span>appr<span class="_ _1"></span>oach<span class="_ _7"> </span>to<span class="_ _8"> </span>joint<span class="_ _7"> </span>classi&#64257;-</div><div class="t m0 x16 h7 y13 ff4 fs4 fc0 sc0 ls0 ws0">cation,<span class="_ _8"> </span>detection<span class="_ _7"> </span>and<span class="_ _8"> </span>semantic<span class="_ _7"> </span>se<span class="_ _1"></span>gmentation<span class="_ _8"> </span>via<span class="_ _7"> </span>a<span class="_ _7"> </span>uni&#64257;ed</div><div class="t m0 x16 h7 y14 ff4 fs4 fc0 sc0 ls0 ws0">ar<span class="_ _1"></span>chitectur<span class="_ _2"></span>e<span class="_"> </span>wher<span class="_ _2"></span>e<span class="_"> </span>the<span class="_"> </span>encoder<span class="_ _b"> </span>is<span class="_"> </span>shar<span class="_ _1"></span>ed<span class="_ _b"> </span>amongst<span class="_"> </span>the<span class="_"> </span>thr<span class="_ _1"></span>ee</div><div class="t m0 x16 h7 y15 ff4 fs4 fc0 sc0 ls0 ws0">tasks.<span class="_ _9"> </span>Our<span class="_ _b"> </span>appr<span class="_ _1"></span>oach<span class="_"> </span>is<span class="_ _b"> </span>very<span class="_ _b"> </span>simple,<span class="_ _b"> </span>can<span class="_"> </span>be<span class="_ _b"> </span>trained<span class="_ _b"> </span>end-to-</div><div class="t m0 x16 h7 y16 ff4 fs4 fc0 sc0 ls0 ws0">end<span class="_ _b"> </span>and<span class="_ _7"> </span>performs<span class="_ _b"> </span>extr<span class="_ _1"></span>emely<span class="_ _b"> </span>well<span class="_ _7"> </span>in<span class="_ _b"> </span>the<span class="_ _7"> </span>c<span class="_ _2"></span>hallenging<span class="_ _b"> </span>KITTI</div><div class="t m0 x16 h7 y17 ff4 fs4 fc0 sc0 ls0 ws0">dataset,<span class="_ _b"> </span>outperforming<span class="_ _b"> </span>the<span class="_ _b"> </span>state-of-the-art<span class="_ _b"> </span>in<span class="_ _b"> </span>the<span class="_ _b"> </span>r<span class="_ _3"></span>oad<span class="_ _b"> </span>seg-</div><div class="t m0 x16 h7 y18 ff4 fs4 fc0 sc0 ls0 ws0">mentation<span class="_ _b"> </span>task.<span class="_ _0"> </span>Our<span class="_ _b"> </span>appr<span class="_ _1"></span>oach<span class="_ _b"> </span>is<span class="_ _b"> </span>also<span class="_ _b"> </span>very<span class="_ _b"> </span>ef<span class="_ _1"></span>&#64257;cient,<span class="_ _7"> </span>taking</div><div class="t m0 x16 h7 y19 ff4 fs4 fc0 sc0 ls0 ws0">less<span class="_"> </span>than<span class="_"> </span>100<span class="_"> </span>ms<span class="_"> </span>to<span class="_"> </span>perform<span class="_"> </span>all<span class="_"> </span>tasks.</div><div class="t m0 x16 h6 y1a ff1 fs1 fc0 sc0 ls0 ws0">1.<span class="_"> </span>Introduction</div><div class="t m0 x15 h8 y1b ff2 fs4 fc0 sc0 ls0 ws0">Current<span class="_ _8"> </span>advances<span class="_ _8"> </span>in<span class="_ _8"> </span>the<span class="_ _8"> </span>&#64257;eld<span class="_ _9"> </span>of<span class="_ _8"> </span>computer<span class="_ _9"> </span>vision<span class="_ _8"> </span>have</div><div class="t m0 x16 h8 y1c ff2 fs4 fc0 sc0 ls0 ws0">made<span class="_"> </span>clear<span class="_"> </span>that<span class="_ _a"> </span>visual<span class="_"> </span>perception<span class="_"> </span>is<span class="_"> </span>going<span class="_ _a"> </span>to<span class="_"> </span>play<span class="_"> </span>a<span class="_"> </span>k<span class="_ _2"></span>ey<span class="_ _a"> </span>role</div><div class="t m0 x16 h8 y1d ff2 fs4 fc0 sc0 ls0 ws0">in<span class="_ _b"> </span>the<span class="_"> </span>dev<span class="_ _2"></span>elopment<span class="_ _b"> </span>of<span class="_"> </span>self-driving<span class="_"> </span>cars.<span class="_ _9"> </span>This<span class="_ _b"> </span>is<span class="_ _b"> </span>mostly<span class="_"> </span>due</div><div class="t m0 x16 h8 y1e ff2 fs4 fc0 sc0 ls0 ws0">to<span class="_ _8"> </span>the<span class="_ _8"> </span>deep<span class="_ _9"> </span>learning<span class="_ _8"> </span>rev<span class="_ _1"></span>olution<span class="_ _9"> </span>which<span class="_ _8"> </span>begun<span class="_ _8"> </span>with<span class="_ _8"> </span>the<span class="_ _9"> </span>in-</div><div class="t m0 x16 h8 y1f ff2 fs4 fc0 sc0 ls0 ws0">troduction<span class="_ _7"> </span>of<span class="_ _b"> </span>AlexNet<span class="_ _b"> </span>in<span class="_ _7"> </span>2012<span class="_ _b"> </span>[<span class="fc1">23</span>].<span class="_ _6"> </span>Since<span class="_ _7"> </span>then,<span class="_ _7"> </span>the<span class="_ _b"> </span>accu-</div><div class="t m0 x16 h8 y20 ff2 fs4 fc0 sc0 ls0 ws0">racy<span class="_ _a"> </span>of<span class="_"> </span>ne<span class="_ _1"></span>w<span class="_"> </span>approaches<span class="_ _a"> </span>has<span class="_"> </span>been<span class="_ _a"> </span>increasing<span class="_"> </span>at<span class="_"> </span>a<span class="_ _a"> </span>vertiginous</div><div class="t m0 x16 h8 y21 ff2 fs4 fc0 sc0 ls0 ws0">rate.<span class="_ _b"> </span>Causes<span class="_ _a"> </span>of<span class="_ _a"> </span>this<span class="_ _a"> </span>are<span class="_ _a"> </span>the<span class="_ _a"> </span>e<span class="_ _1"></span>xistence<span class="_ _a"> </span>of<span class="_ _a"> </span>more<span class="_ _a"> </span>data,<span class="_ _a"> </span>increased</div><div class="t m0 x16 h8 y22 ff2 fs4 fc0 sc0 ls0 ws0">computation<span class="_ _a"> </span>po<span class="_ _1"></span>wer<span class="_ _a"> </span>and<span class="_ _a"> </span>algorithmic<span class="_ _a"> </span>de<span class="_ _2"></span>velopments.<span class="_ _b"> </span>The<span class="_ _c"> </span>cur-</div><div class="t m0 x16 h8 y23 ff2 fs4 fc0 sc0 ls0 ws0">rent<span class="_"> </span>trend<span class="_ _b"> </span>is<span class="_ _b"> </span>to<span class="_"> </span>create<span class="_ _b"> </span>deeper<span class="_"> </span>networks<span class="_"> </span>with<span class="_ _b"> </span>as<span class="_ _b"> </span>man<span class="_ _1"></span>y<span class="_ _b"> </span>layers</div><div class="t m0 x16 h8 y24 ff2 fs4 fc0 sc0 ls0 ws0">as<span class="_"> </span>possible<span class="_"> </span>[<span class="fc1">17</span>].</div><div class="t m0 x15 h8 y25 ff2 fs4 fc0 sc0 ls0 ws0">While<span class="_ _a"> </span>performance<span class="_ _a"> </span>is<span class="_ _a"> </span>extremely<span class="_ _c"> </span>high,<span class="_"> </span>when<span class="_ _c"> </span>dealing<span class="_ _a"> </span>with</div><div class="t m0 x16 h8 y26 ff2 fs4 fc0 sc0 ls0 ws0">real-world<span class="_ _8"> </span>applications,<span class="_ _9"> </span>running<span class="_ _8"> </span>times<span class="_ _8"> </span>become<span class="_ _8"> </span>important.</div><div class="t m0 x16 h8 y27 ff2 fs4 fc0 sc0 ls0 ws0">New<span class="_ _c"> </span>hardware<span class="_ _a"> </span>accelerators<span class="_"> </span>as<span class="_ _c"> </span>well<span class="_"> </span>as<span class="_ _c"> </span>compression,<span class="_"> </span>reduced</div><div class="t m0 x16 h8 y28 ff2 fs4 fc0 sc0 ls0 ws0">precision<span class="_ _8"> </span>and<span class="_ _8"> </span>distillation<span class="_ _8"> </span>methods<span class="_ _9"> </span>hav<span class="_ _2"></span>e<span class="_ _8"> </span>been<span class="_ _8"> </span>exploited<span class="_ _8"> </span>to</div><div class="t m0 x16 h8 y29 ff2 fs4 fc0 sc0 ls0 ws0">speed<span class="_"> </span>up<span class="_"> </span>current<span class="_"> </span>networks.</div><div class="t m0 x15 h8 y2a ff2 fs4 fc0 sc0 ls0 ws0">In<span class="_ _9"> </span>this<span class="_ _9"> </span>paper<span class="_ _9"> </span>we<span class="_ _9"> </span>take<span class="_ _9"> </span>an<span class="_ _9"> </span>alternativ<span class="_ _2"></span>e<span class="_ _9"> </span>approach<span class="_ _9"> </span>and<span class="_ _9"> </span>de-</div><div class="t m0 x16 h8 y2b ff2 fs4 fc0 sc0 ls0 ws0">sign<span class="_ _9"> </span>a<span class="_ _9"> </span>network<span class="_ _0"> </span>architecture<span class="_ _9"> </span>that<span class="_ _9"> </span>can<span class="_ _9"> </span>very<span class="_ _9"> </span>ef&#64257;ciently<span class="_ _9"> </span>per-</div><div class="t m0 x16 h8 y2c ff2 fs4 fc0 sc0 ls0 ws0">form<span class="_ _a"> </span>classi&#64257;cation,<span class="_"> </span>detection<span class="_ _a"> </span>and<span class="_ _a"> </span>semantic<span class="_"> </span>se<span class="_ _1"></span>gmentation<span class="_"> </span>si-</div><div class="t m0 x16 h8 y2d ff2 fs4 fc0 sc0 ls0 ws0">multaneously<span class="_ _3"></span>.<span class="_ _6"> </span>This<span class="_ _b"> </span>is<span class="_ _7"> </span>done<span class="_ _b"> </span>by<span class="_ _7"> </span>incorporating<span class="_ _b"> </span>all<span class="_ _7"> </span>three<span class="_ _b"> </span>task</div><div class="t m0 x17 h8 y2e ff2 fs4 fc0 sc0 ls0 ws0">Figure<span class="_ _8"> </span>1:<span class="_ _d"> </span>Our<span class="_ _8"> </span>goal:<span class="_ _d"> </span>Solving<span class="_ _8"> </span>street<span class="_ _8"> </span>classi&#64257;cation,<span class="_ _0"> </span>vehicle</div><div class="t m0 x17 h8 y2f ff2 fs4 fc0 sc0 ls0 ws0">detection<span class="_"> </span>and<span class="_"> </span>road<span class="_"> </span>segmentation<span class="_"> </span>in<span class="_"> </span>one<span class="_"> </span>forward<span class="_"> </span>pass.</div><div class="t m0 x17 h8 y30 ff2 fs4 fc0 sc0 ls0 ws0">into<span class="_ _7"> </span>a<span class="_ _b"> </span>uni&#64257;ed<span class="_ _7"> </span>encoder-decoder<span class="_ _b"> </span>architecture.<span class="_ _6"> </span>W<span class="_ _1"></span>e<span class="_ _7"> </span>name<span class="_ _b"> </span>our</div><div class="t m0 x17 h8 y31 ff2 fs4 fc0 sc0 ls0 ws0">approach<span class="_ _7"> </span>MultiNet.<span class="_ _d"> </span>The<span class="_ _7"> </span>encoder<span class="_ _8"> </span>consists<span class="_ _7"> </span>of<span class="_ _7"> </span>the<span class="_ _7"> </span>con<span class="_ _2"></span>volu-</div><div class="t m0 x17 h8 y32 ff2 fs4 fc0 sc0 ls0 ws0">tion<span class="_ _8"> </span>and<span class="_ _9"> </span>pooling<span class="_ _9"> </span>layers<span class="_ _8"> </span>from<span class="_ _9"> </span>the<span class="_ _9"> </span>VGG<span class="_ _8"> </span>network<span class="_ _8"> </span>[<span class="fc1">45</span>]<span class="_ _9"> </span>and</div><div class="t m0 x17 h8 y33 ff2 fs4 fc0 sc0 ls0 ws0">is<span class="_ _b"> </span>shared<span class="_ _b"> </span>among<span class="_ _7"> </span>all<span class="_ _b"> </span>tasks.<span class="_ _0"> </span>Those<span class="_ _b"> </span>features<span class="_ _7"> </span>are<span class="_ _b"> </span>then<span class="_ _b"> </span>utilized</div><div class="t m0 x17 h8 y34 ff2 fs4 fc0 sc0 ls0 ws0">by<span class="_ _8"> </span>task-speci&#64257;c<span class="_ _8"> </span>decoders,<span class="_ _9"> </span>which<span class="_ _8"> </span>produce<span class="_ _8"> </span>their<span class="_ _8"> </span>outputs<span class="_ _8"> </span>in</div><div class="t m0 x17 h8 y35 ff2 fs4 fc0 sc0 ls0 ws0">real-time.<span class="_ _7"> </span>In<span class="_"> </span>particular<span class="_ _1"></span>,<span class="_"> </span>the<span class="_ _b"> </span>detection<span class="_"> </span>decoder<span class="_"> </span>combines<span class="_"> </span>the</div><div class="t m0 x17 h8 y36 ff2 fs4 fc0 sc0 ls0 ws0">fast<span class="_ _a"> </span>regression<span class="_"> </span>design<span class="_ _a"> </span>introduced<span class="_"> </span>in<span class="_ _a"> </span>Y<span class="_ _5"></span>olo<span class="_"> </span>[<span class="fc1">38</span>]<span class="_"> </span>with<span class="_ _a"> </span>the<span class="_"> </span>size-</div><div class="t m0 x17 h8 y37 ff2 fs4 fc0 sc0 ls0 ws0">adjusting<span class="_ _a"> </span>R<span class="_ _2"></span>OI-Pooling<span class="_ _a"> </span>of<span class="_"> </span>F<span class="_ _2"></span>ast-RCNN<span class="_ _a"> </span>[<span class="fc1">14</span>],<span class="_"> </span>achie<span class="_ _1"></span>ving<span class="_"> </span>a<span class="_ _a"> </span>bet-</div><div class="t m0 x17 h8 y38 ff2 fs4 fc0 sc0 ls0 ws0">ter<span class="_"> </span>speed-accuracy<span class="_"> </span>ratio.</div><div class="t m0 x18 h8 y39 ff2 fs4 fc0 sc0 ls0 ws0">W<span class="_ _3"></span>e<span class="_"> </span>demonstrate<span class="_ _b"> </span>the<span class="_"> </span>ef<span class="_ _2"></span>fectiv<span class="_ _1"></span>eness<span class="_ _b"> </span>of<span class="_"> </span>our<span class="_"> </span>approach<span class="_ _b"> </span>in<span class="_"> </span>the</div><div class="t m0 x17 h8 y3a ff2 fs4 fc0 sc0 ls0 ws0">challenging<span class="_ _b"> </span>KITTI<span class="_ _7"> </span>benchmark<span class="_ _b"> </span>[<span class="fc1">13</span>]<span class="_ _7"> </span>and<span class="_ _b"> </span>show<span class="_ _b"> </span>state-of-the-</div><div class="t m0 x17 h8 y3b ff2 fs4 fc0 sc0 ls0 ws0">art<span class="_ _0"> </span>performance<span class="_ _0"> </span>in<span class="_ _0"> </span>road<span class="_ _e"> </span>segmentation.<span class="_ _f"> </span>Importantly<span class="_ _3"></span>,<span class="_ _6"> </span>our</div><div class="t m0 x17 h8 y3c ff2 fs4 fc0 sc0 ls0 ws0">R<span class="_ _1"></span>OI-Pooling<span class="_ _b"> </span>implementation<span class="_ _b"> </span>can<span class="_"> </span>signi&#64257;cantly<span class="_ _b"> </span>improv<span class="_ _1"></span>e<span class="_ _b"> </span>de-</div><div class="t m0 x17 h8 y3d ff2 fs4 fc0 sc0 ls0 ws0">tection<span class="_ _b"> </span>performance<span class="_ _b"> </span>without<span class="_ _7"> </span>requiring<span class="_ _b"> </span>an<span class="_ _b"> </span>explicit<span class="_ _b"> </span>proposal</div><div class="t m0 x17 h8 y3e ff2 fs4 fc0 sc0 ls0 ws0">generation<span class="_ _8"> </span>network.<span class="_ _10"> </span>This<span class="_ _8"> </span>giv<span class="_ _1"></span>es<span class="_ _9"> </span>our<span class="_ _8"> </span>decoder<span class="_ _8"> </span>a<span class="_ _8"> </span>signi&#64257;cant</div><div class="t m0 x17 h8 y3f ff2 fs4 fc0 sc0 ls0 ws0">speed<span class="_ _b"> </span>adv<span class="_ _1"></span>antage<span class="_ _b"> </span>compared<span class="_ _b"> </span>to<span class="_"> </span>Faster-RCNN.<span class="_"> </span>Our<span class="_ _b"> </span>approach</div><div class="t m0 x17 h8 y40 ff2 fs4 fc0 sc0 ls0 ws0">is<span class="_"> </span>able<span class="_"> </span>to<span class="_"> </span>bene&#64257;t<span class="_"> </span>from<span class="_"> </span>sharing<span class="_ _a"> </span>computations,<span class="_"> </span>allowing<span class="_"> </span>us<span class="_"> </span>to</div><div class="t m0 x17 h8 y41 ff2 fs4 fc0 sc0 ls0 ws0">perform<span class="_"> </span>inference<span class="_"> </span>in<span class="_"> </span>less<span class="_"> </span>than<span class="_"> </span>100<span class="_"> </span>ms<span class="_"> </span>for<span class="_"> </span>all<span class="_"> </span>tasks.</div><div class="t m0 x17 h6 y42 ff1 fs1 fc0 sc0 ls0 ws0">2.<span class="_"> </span>Related<span class="_"> </span>W<span class="_ _3"></span>ork</div><div class="t m0 x18 h8 y2a ff2 fs4 fc0 sc0 ls0 ws0">In<span class="_"> </span>this<span class="_"> </span>section<span class="_"> </span>we<span class="_"> </span>re<span class="_ _2"></span>vie<span class="_ _2"></span>w<span class="_"> </span>current<span class="_"> </span>approaches<span class="_"> </span>to<span class="_"> </span>the<span class="_"> </span>tasks</div><div class="t m0 x17 h8 y2b ff2 fs4 fc0 sc0 ls0 ws0">that<span class="_ _b"> </span>MultiNet<span class="_ _7"> </span>tackles,<span class="_ _7"> </span>i.e.,<span class="_ _7"> </span>detection,<span class="_ _7"> </span>classi&#64257;cation<span class="_ _b"> </span>and<span class="_ _7"> </span>se-</div><div class="t m0 x17 h8 y2c ff2 fs4 fc0 sc0 ls0 ws0">mantic<span class="_"> </span>se<span class="_ _1"></span>gmentation.<span class="_ _7"> </span>W<span class="_ _5"></span>e<span class="_"> </span>focus<span class="_ _a"> </span>our<span class="_"> </span>attention<span class="_ _a"> </span>on<span class="_"> </span>deep<span class="_ _a"> </span>learn-</div><div class="t m0 x17 h8 y2d ff2 fs4 fc0 sc0 ls0 ws0">ing<span class="_"> </span>based<span class="_"> </span>approaches.</div><div class="t m0 x19 h8 y43 ff2 fs4 fc0 sc0 ls0 ws0">1</div><div class="t m1 x1a h9 y44 ff5 fs5 fc2 sc0 ls0 ws0">arXiv:1612.07695v1 [cs.CV] 22 Dec 2016</div><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a><a class="l" rel='nofollow' onclick='return false;'><div class="d m2"></div></a></div><div class="pi" data-data='{"ctm":[1.568627,0.000000,0.000000,1.568627,0.000000,0.000000]}'></div></div> </body> </html>
评论
    相关推荐