ÿþ<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"> <html> <head> <title>Ping Tan's Research</title> <style type="text/css"> .style3 { font-size: 14pt; font-weight: bold; } </style> </head> <body> <table id="PersonalInfo" border="0" align=center style="FONT-SIZE: 14pt; font-family: 'Times New Roman'" width="600"> <tr> <td rowspan="5" width="20%"> <img src = "Photo_small.jpg"/> </td> <td rowspan="5" width="5%"> </td> <td style="width: 27%"> <span style="FONT-SIZE: 18pt"> Ping TAN </span> </td> <td colspan = "2" style="width: 40%"> <img src = "Name_small.bmp" /> </td> </tr> <tr> <td colspan = "3" width="75%"> <span style="font-size: 14pt"> Assistant Professor </span> </td> </tr> <tr> <td colspan="3" width="75%"> Department of Electrical &amp; Computer Engineering</td> </tr> <tr> <td colspan="3" width="75%"> National University of Singapore</td> </tr> <tr> <td colspan="1" style="width: 27%; height: 4px;"> Email: </td> <td colspan="2" width="65%" style="height: 4px"> <img src =Email.png > </td> </tr> </table> <table id="Navigation" border="0" align = "center" style="FONT-SIZE: 14pt; font-family: 'Times New Roman'" width="800"> <tr> <td width="15%" > <p align="center"> <a href="Index.htm"> Home</a></p> </td> <td width="16%"> <p align="center"> <a href="Research.htm"> Research </a></p> </td> <td width = "16%"> <p align="center"> <a href="Publication.htm"> Publication </a></p> </td> <td width="16%"> <p align="center"> <a href="Contact.htm"> Contact </a></p> </td> <td width="15%"> <p align="center"> <a href="Misc.htm"> Misc </a></p> </td> </tr> </table> <br /> <table id="News" border="0" align = "center" style="FONT-SIZE: 14pt; font-family: 'Times New Roman'"; width="800"> </table> <table id="Research" border="0" align = "center" style="FONT-SIZE: 12pt; font-family: 'Times New Roman'"; width="800"> <tr> <td colspan = "2" class="style3"> Data-driven Media Synthesis</td> </tr> <tr> <td colspan = "2" style="text-align: justify" > While professional artists can quickly create high quality images and animations, most of ordinary people can hardly do that. We are exploiting the large amount of internet data to design algorithms in a data-driven approach to simplify visual media creation.</td> </tr> <tr> <td align = "center"> <img border = "0" src ="research/sketch2photo.jpg" style="height: 110px; width: 283px;" /> </td> <td style="text-align: justify"> <i>Sketch2Photo</i> <br /> A picture is worth a thousand words, and not surprising, people often use pictures and animations to convey ideas and stories. However, the ability for ordinary people to quickly sketch-up realistic image-based media remains elusive. We developed methods to automatically convert an annotated freehand doodle into a realistic picture (<a href="Papers/sigasia09_sketch2photo.pdf">SIGGRAPH Asia 2009</a>) or a comic strip (<a href="Papers/poseshop.pdf">Technique Report</a>). Our results were generated by stitching multiple photographs in agreement with the sketch and text labels. These images were found by automatically searching the Internet. More details can be found in the paper and the <a href="Projects/Sketch2Photo/index.htm">project page</a>. <br /> <br /> </td> </tr> <tr> <td align = "center"><img border = "0" src ="research/colorization.JPG" style="height: 110px;" /> <td style="text-align: justify"> <i>Semantic Colorization</i> <br /> Image colorization can bring a grayscale photo to life, but often demands extensive user interaction. We developed a method to turn a grayscle image into color according to suitable reference example images automatically searched from the Internet. Our method only requires the user to provide a semantic text label and segmentation cues for major foreground objects in the scene. More detail can be found in the paper (<a href="Papers/sigasia11.pdf">SIGGRAPH Asia 2011</a>). </td> </tr> <tr> <td colspan = "2" class="style3"> Image-based Modleing</td> </tr> <tr> <td colspan = "2" style="text-align: justify"> Realistic 3D models are important for computer graphics applications. Most existing modeling systems still heavily rely on user interactions to manually model details. We are working towards the goal of automatic creation of realistic 3D models from images of real objects. </tr> <tr> <td align = "center"><img border = "0" src ="research/treemodeling.JPG" style="height: 110px;" /> <td style="text-align: justify"> <i>Image-based Tree Modeling</i> <br /> Trees are everywhere and are difficult to model in a realistic way. Previous methods mainly rely on 'shape grammars' and 'geometric rules' which are difficult to use and the results are difficult to control. We developed methods to easily recover realistic 3D models from 2D images. This approach was applied to small plants (<a href="Papers/siggraph06.pdf">SIGGRAPH 2006</a>) and trees (<a href="Papers/siggraph07.pdf">SIGGRAPH 2007</a>). We also developed a method for the extreme case when only one input image is available (<a href="Papers/sigasia08_tree.pdf">SIGGRAPH Asia 2008</a>). Video demos and 3D models can be found at the <a href="Projects/ImageBasedModeling/index.htm">project page</a>. <br /> <br /> </td> </tr> <tr> <td align = "center"><img border = "0" src ="research/architecturemodeling2.jpg" style="height: 110px;" /> </td> <td style="text-align: justify"> <i>Image-based Architecture Modeling</i> <br /> There is a strong demand for the photo-realistic modeling of cities for games, movies and map services (e.g.Google Earth). However, manual modeling is tedious and time consuming. Previous image-based modeling works often use aerial photographs which give little architecture details. We applied the image-based modeling approach with street level images to model buildings at the scale of a city block (<a href="Papers/sigasia08_street.pdf">SIGGRAPH Asia 2008</a>) or a single building (<a href="Papers/sigasia09_building.pdf">SIGGRAPH Asia 2009</a>). Video demos and 3D models can be found at the <a href="Projects/ImageBasedModeling/index.htm">project page</a>. <br /> </td> </tr> <tr> <td colspan = "2" class="style3"> 3D Reconstruction & Visual SLAM</td> </tr> <tr> <td colspan = "2" style="text-align: justify"> Real-time 3D reconstruction can be used to facilitate autonmous navigation of robots and unmaned vehicles. We are working to improve the structure-from-motion algorithm, and to apply it for simultaneous localization and mapping (SLAM) applications.</td> </tr> <tr> <td align = "center"><img border = "0" src ="research/sfm.png" style="height: 110px;" /></td> <td style="text-align: justify"> <i>Structure-from-Motion</i> <br /> Feature matching errors cause ambiguous epipolar geometry, which leads to incorrect 3D reconstructions. We introduced a numeric measure for the correctness of a given 3D reconstruction, and designed an algorithm to quickly upgrade an incorrect reconstruction to a correct one (<a href="Papers/cvpr12_sfm.pdf">CVPR 2012</a>). <br /> <br /> </td> </tr> <tr> <td align = "center"><img border = "0" src ="research/slam.png" style="height: 110px;" /></td> <td style="text-align: justify"> <i>Visual SLAM</i> <br /> Many visual SLAM algorithms have been designed for static scenes, but real world is full of dynamic objects. We studied the visual SLAM problem in dynamic scenes with multiple independently moving cameras (<a href="Papers/pami12_slam.pdf">PAMI 2012</a>). Video demos can be found at the <a href="Projects/SLAM/index.htm">project page</a>. <br /> <br /> </td> </tr> <tr> <td colspan = "2" class="style3"> Physically-based Vision</td> </tr> <tr> <td colspan = "2" style="text-align: justify"> An image of an object is determined through complex interactions between its reflectance, shape, sourring environment and the imaging process. We are seeking to invert this process and recover scene information. </td> </tr> <tr> <td align = "center" > <img border = "0" src ="research/symmetries.jpg" style="width: 231px; height: 110px;" /> </td> <td style="text-align: justify"> <i>Reflectance Symmetries</i> <br /> When recovering shape from diffuse shading we are faced with an intrinsic shape/lighting ambiguity. We showed that almost any additive specular reflection component resolves this ambiguity. The basic idea is to exploit symmetries (reciprocity and isotropy) in the BRDF, which induce joint constraints on shape, lighting, and viewpoint. These constraints can be described on the Gaussian sphere (<a href="Papers/cvpr07.pdf">CVPR2007</a>) or the real projective plane (<a href="Papers/cvpr09_projective.pdf">CVPR2009</a>). For a more comprehensive summary, please refer to the journal version (<a href="Papers/pami10_geometry.pdf">PAMI2011</a>). </td> </tr> <tr> <td align = "center"> <img border = "0" src ="research/photometric.jpg" style="width: 234px; height: 110px;" /> </td> <td style="text-align: justify"> <i>Photometric Stereo</i> <br /> Local surface orientations can be precisely deteremined from multiple images captured by a fixed camera under variant known illumination conditions. We developed algorithms to automatically determine the lighting conditions for smooth surfaces (<a href="Papers/cvpr10.pdf">CVPR2010</a>) and discrete surfaces (<a href="Papers/eccv10.pdf">ECCV2010</a>). We also reconstructed the subpixel level geometric structures on rough surfaces (<a href="Papers/pami07.pdf">PAMI2008</a>) and applied photometric stereo to online images (<a href="Papers/cvpr09_outdoorps.pdf">CVPR2009</a>). </td> </tr> <tr> <td align = "center" > <img border = "0" src ="research/deblur.jpg" style="height: 100px; width: 251px" /> </td> <td style="text-align: justify"> <i>Spatially Variant Blur</i> <br /> A common assumption in existing motion deblurring algorithms is that all pixels undergo with the same amount of blur. However, real images often have spatially variant blur. We developed a mathematic model to describe spatially variant blur and applied to make a blurry image clear (<a href="Papers/pami10_deblur.pdf">PAMI2011</a>). </td> </tr> <tr> <td align = "center"> <img border = "0" src ="research/multires.bmp" style="height: 110px; width: 203px" /> </td> <td style="text-align: justify"> <i>Resolution Dependent Reflectance Modeling</i> <br /> The reflectance of a surface depends upon the resolution at which it is imaged. We propose a framework for real-time rendering of multiresolution reflectance that is suitable for a wide range of parametric BRDFs (<a href="Papers/tvcg07.pdf">TVCG08</a>). </td> </tr> <tr> <td align = "center"> <img border = "0" src ="research/separation.jpg" style="height: 95px; width: 266px" /> </td> <td style="text-align: justify"> <i>Reflectance Separation</i> <br /> Surface reflectance can be separated into diffuse and specular components, each with different physical properties and might be applied to vision problems in different manners. We developed methods to achieve this separation of regular surfaces (<a href="Papers/iccv03.pdf">ICCV2003</a>) and texture surfaces (<a href="Papers/cvpr06.pdf">CVPR2006</a>). On the other hand, an image might also be separated into two intrinsic components: reflectance and shading. We also developed algorithms for this separation (<a href="Papers/cvpr08.pdf">CVPR2008</a>). </td> </tr> <tr> <td colspan = "2"> <br /> <br /> <br /> </td> </tr> <tr> <td colspan = "2"></td> </tr> <tr> <td colspan = "2">Last modified in Apr 2012</td> </tr> </table> <table id="Table1" border="0" align = "center" style="FONT-SIZE: 14pt; font-family: 'Times New Roman'"; width="800" height="400"> </table> </body> </html>