<div dir="ltr"><span class="gmail-im" style="color:rgb(80,0,80)"><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Dear colleagues,<span class="gmail-m_-1452529743735516927apple-converted-space"> </span></span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black"> </span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">We cordially invite you to attend the ICCV’19 tutorial on<span class="gmail-m_-1452529743735516927apple-converted-space"> </span><b>Global Optimization for Geometric Understanding with Provable Guarantees.</b></span><span style="font-size:12pt;font-family:Arial,sans-serif"></span><u></u><u></u></p><p class="MsoNormal"> <u></u><u></u></p></span><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">The tutorial will be held in the morning of Sunday</span><span style="font-size:12pt;font-family:Arial,sans-serif">,<span style="color:black"> October 27 in room #402 (first day of ICCV!).</span></span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif"> </span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">We will discuss novel optimization tools for </span><span style="font-size:12pt;font-family:Arial,sans-serif">robust <span style="color:black">3D vision</span> (particularly important for safety critical robotics applications)<span style="color:black">, provide a snapshot of the state of the art, and outline a number of open problems.</span> We trust this will be a useful event<span style="color:black">, with multiple talks, hands-on examples, and open discussion.</span> <span style="color:black">Please find more details below:</span></span><u></u><u></u></p><span class="gmail-im" style="color:rgb(80,0,80)"><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black"> </span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">------------------------------------------------------------------------------------</span><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Global Optimization for Geometric Understanding with Provable Guarantees</span></b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black"> (tutorial)</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">International Conference on Computer Vision (ICCV)</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Date: October 27, 2019, Seoul</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Website: <a href="http://globaloptimization-iccv2019.mit.edu/" target="_blank">http://globalOptimization-ICCV2019.mit.edu</a></span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Room: #402</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black"> </span><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">OVERVIEW:</span></b><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">----------------</span></b><u></u><u></u></p><p style="margin:0in 0in 0.0001pt;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">This tutorial aims to give an in-depth introduction to global optimization tools, including convex and semidefinite relaxations, applied to 3D vision problems. The first goal of the tutorial is to motivate the need for global solvers by providing real-world examples where the lack of robustness results from the difficulty in solving large optimization problems to optimality. The second goal is to provide the attendees with basic mathematical and algorithmic concepts and survey important recent advances in the area. The third goal is to outline several open research avenues: global optimization has an enormous untapped potential and it is hoped that this tutorial will inspire researchers to use modern optimization tools to solve several outstanding challenges in geometric vision. </span><u></u><u></u></p><p style="margin:0in 0in 0.0001pt;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black"> </span><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">DETAILED DESCRIPTION:</span></b><u></u><u></u></p></span><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">------------------------------------<br></span></b><span class="gmail-im" style="color:rgb(80,0,80)"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Understanding the geometry of a scene from camera observations is a key requirement from many applications, ranging from autonomous vehicles (e.g., self-driving cars, consumer drones) to Virtual and Augmented Reality. Indeed, geometric understanding encompasses several core topics in computer vision and robotics, including Structure from Motion and SLAM, point cloud registration and object pose estimation, single-view and two-view geometry, among many others. Despite the maturity of the algorithms developed for geometric understanding, both researchers and practitioners are well aware that the presence of large noise, outliers, and missing data makes modern pipelines brittle when operating in the wild. An incorrect understanding of the geometry of the scene can negatively affect the user experience in consumer applications, and may put human life at risk in safety-critical applications, such as self-driving cars.</span><u></u><u></u></span></p><p style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">One of the main causes of the fragility of geometric understanding is the fact that modern pipelines typically trade-off performance (robustness, optimality, guarantees) for computational efficiency. Indeed, many optimization problems underlying geometric understanding are intractable due to non-convexity or due to their combinatorial nature (e.g., measurement selection to reject potential outliers). Several algorithms obtain fast solutions using local iterative nonlinear solvers or heuristic approaches. While heuristics and local solvers tend to work well in the low-noise low-outlier regime, they are prone to fail in more challenging real-world conditions.</span><u></u><u></u></p><p style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Global optimization and convex relaxation techniques have been recently shown to provide an effective tool to tame the complexity of geometric understanding while enabling efficient solutions. A sequence of papers have debunked two common misconceptions behind these methods. The first misconception is that a convex relaxation always entails computing an “approximate solution”: indeed recent papers have shown that, in certain regimes, one can design convex relaxations that (provably) solve the original problem exactly. The second misconception is that many of these relaxations, in particular semidefinite relaxations, are slow in practice: modern solvers and theory instead show that one can solve several classes of such relaxations very efficiently. A growing body of work in vision and robotics demonstrates that these tools can indeed solve in a provably optimal manner problems including rotation averaging and SLAM, registration, and combinatorial problems involving outlier rejection.</span><u></u><u></u></p><span class="gmail-im" style="color:rgb(80,0,80)"><p style="margin:0in 0in 0.0001pt;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">This tutorial aims to give an in-depth introduction to global optimization tools, including convex and semidefinite relaxations. The first goal of the tutorial is to motivate the need for global solvers by providing real-world examples where the lack of robustness results from the difficulty in solving large optimization problems to optimality. The second goal is to provide the attendees with basic mathematical and algorithmic concepts, and survey important recent advances in the area. The third goal is to outline several open research avenues: global optimization has an enormous untapped potential and it is hoped that this tutorial will inspire researchers to use modern optimization tools to solve several outstanding challenges in geometric vision. </span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">  </span><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">ORGANIZING COMMITTEE:</span></b><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">-------------------------------------</span></b><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">- Luca Carlone, Massachusetts Institute of Technology</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">- Tat-Jun Chin, The University of Adelaide</span><u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">- Anders Eriksson, University of Queensland</span><u></u><u></u></p></span><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif">- Heng Yang<span style="color:black">, Massachusetts Institute of Technology</span></span><u></u><u></u></p><span class="gmail-im" style="color:rgb(80,0,80)"><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">- Fredrik Kahl, Chalmers University of Technology</span><u></u><u></u></p><p class="MsoNormal"> <u></u><u></u></p><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">  </span><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">FURTHER INFORMATION:</span></b><u></u><u></u></p><p class="MsoNormal"><b><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">-----------------------------------</span></b><u></u><u></u></p></span><p class="MsoNormal"><span style="font-size:12pt;font-family:Arial,sans-serif;color:black">Please send any questions to<span class="gmail-m_-1452529743735516927apple-converted-space"> </span>Luca Carlone (<a href="mailto:lcarlone@mit.edu" target="_blank"><span style="color:rgb(113,60,86)">lcarlone@mit.edu</span></a>), Tat-Jun Chin (<a href="mailto:tat-jun.chin@adelaide.edu.au" target="_blank">tat-jun.chin@adelaide.edu.au</a>), </span><span style="font-size:12pt;font-family:Arial,sans-serif">Heng Yang<span style="color:black"> (<a href="mailto:hankyang@mit.edu" target="_blank">hankyang@mit.edu</a>), or Anders Eriksson (<a href="mailto:a.eriksson@uq.edu.au" target="_blank">a.eriksson@uq.edu.au</a>).</span></span></p></div>