1<html> 2 <head> 3 <title>DeepPruner</title> 4 <link rel="stylesheet" type="text/css" href="styles.css" /> 5 <link rel="stylesheet" href="https://cdn.rawgit.com/jpswalsh/academicons/master/css/academicons.min.css"> 6 <link href="https://fonts.googleapis.com/css?family=Roboto:light,normal" rel="stylesheet"> 7 <link href="https://fonts.googleapis.com/css?family=Oregano" rel="stylesheet"> 8 <link href="https://fonts.googleapis.com/css?family=Inconsolata" rel="stylesheet"> 9
9<script src="https://kit.fontawesome.com/2ff36a40d1.js"></script>
9 10
10<script type="text/javascript" src="../js/modernizr-1.5.min.js"></script>
10 11
11<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.1.0/jquery.min.js"></script>
11 12
12<script src="https://code.jquery.com/jquery-1.10.2.js"></script>
12 13 </head> 14 15 <body> 16 <div id="content"> 17 <h1>DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch</h1> 18 <div style="text-align: center"> 19 <span class="author"><a href="https://shivamduggal4.github.io/" target="_blank">Shivam Duggal</a><sup>1</sup></span> 20 <span class="author"><a href="http://www.cs.toronto.edu/~slwang/" target="_blank">Shenlong Wang</a><sup>1,2</sup></span> 21 <span class="author"><a href="http://people.csail.mit.edu/weichium/" target="_blank">Wei-Chiu Ma</a><sup>1,3</sup></span> 22 <span class="author"><a href="" target="_blank">Rui Hu</a><sup>1</sup></span> 23 <span class="author" style="padding-right: 0"><a href="http://www.cs.toronto.edu/~urtasun/" target="_blank">Raquel Urtasun</a><sup>1,3</sup></span> 24 </div> 25 <div class="affil"> 26 <span style="padding-right: 2em"><sup>1</sup>Uber Advanced Technologies Group</span> 27 <span style="padding-right: 0em"><sup>2</sup>University of Toronto</span> 28 <br><sup>3</sup>Massachusetts Institute of Technology 29 </div> 30 <div class="venue">IEEE International Conference on Computer Vision (<b>ICCV</b>), 2019</div> 31 <br> 32 <!-- <div class="highlight"><b>1st place on KITTI Scene Flow benchmark</b></div> --> 33 34 <div style="text-align: center;"> 35 <img src="images/deeppruner_vis.png" width=1000><br> 36 </div> 37 38 <h2>Abstract</h2> 39 <p> 40 <b>Our goal is to significantly speed up the runtime of current state-of-the-art stereo algorithms to enable real-time inference.</b> Towards this goal, we developed a Differentiable PatchMatch module that allows us to discard most disparities without requiring full cost volume evaluation. We then exploit this representation to learn which range to prune for each pixel. By progressively reducing the search space and effectively propagating such information, we are able to efficiently compute the cost volume for high likelihood hypotheses and achieve savings in both memory and computation. Finally, an image guided refinement module is exploited to further improve the performance. Since all our components are differentiable, the full network can be trained end-to-end. <b>Our experiments show that our method achieves competitive results on KITTI and SceneFlow datasets while running in real-time at 62ms.</b> 41 </p> 42 <br> 43 44 <h2>Overview of DeepPruner</h2> 45 <!-- <div style="text-align: center;"> 46 <img src="poster.png" width="1000"></img> 47 </div> --> 48 49 <div style="text-align: center;"> 50 <img src="images/deeppruner.png" width="1000"></img> 51 </div> 52 53 <h2>Overview of Patch Match (Barnes et al.) & Differentiable Patch Match</h2> 54 55 <div style="text-align: center;"> 56 <table> 57 <tr> 58 <td><img src="images/PM_overview.png" width="500"/></td> 59 <td><img src="images/DPM_overview.png" width="500"/></td> 60 </tr> 61 </table> 62 </div> 63 64 <h2>Quantitative Results</h2> 65 66 <div style="text-align: center;"> 67 <img src="images/sceneflow_result.png" width="900"/><br><br><br> 68 <table align="center"> 69 <tr> 70 <td><img src="images/kitti_result.png" width="450"/></td> 71 <td><img src="images/rob_result.png" width="450"/></td> 72 </tr> 73 </table> 74 </div> 75 76 <h2>Qualitative Results</h2> 77 78 <div style="text-align: center;"> 79 <img src="images/kitti_vis.png" width="900"/><br><br><br> 80 <table align="center"> 81 <tr> 82 <td><img src="images/sceneflow_vis.png" width="450"/></td> 83 <td><img src="images/rob_vis.png" width="450"/></td> 84 </tr> 85 </table> 86 </div> 87 88 89 <br> 90 91 <h2>Citation</h2> 92 <div class="paper"> 93 <div class="paper_pic"><img src="images/paper_pic.png" width=200></div> 94 <div class="paper_content"> 95 <h3><b>DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch</b></h3> 96 <h4><span style="font-weight:normal;">Shivam Duggal, Shenlong Wang, Wei-Chiu Ma, Rui Hu, Raquel Urtasun</span></h3> 97 <h4>Supplementary <a href="supplementary.pdf" target="_blank"><i class="fa fa-file-pdf-o fa-lg" aria-hidden="true"></i></a> | 98 arXiv preprint <a href="https://arxiv.org/pdf/1909.05845.pdf" target="_blank"><i class="fa fa-file-pdf-o fa-lg" aria-hidden="true"></i></a> | 99 Code <a href="https://github.com/uber-research/DeepPruner" target="_blank"><i class="fab fa-github fa-lg"></i></a></h4> 100 <h3>BibTex</h3> 101 <div class="bibtex"> 102 @InProceedings{Duggal_2019_ICCV,<br> 103 author = {Duggal, Shivam and Wang, Shenlong and Ma, Wei-Chiu and Hu, Rui and Urtasun, Raquel},<br> 104 title = {DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMat
104ch},<br> 105 booktitle = {The IEEE International Conference on Computer Vision (ICCV)},<br> 106 month = {October},<br> 107 year = {2019}<br> 108 } 109 </div> 110 </div> 111 </div> 112 </div> 113 </body> 114</html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.