1<!DOCTYPE HTML> 2<html lang="en"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8"> 3 4 <title>Yunfan Ye's Homepage</title> 5 6 <meta name="author" content="Yunfan Ye"> 7 <meta name="viewport" content="width=device-width, initial-scale=1"> 8 9 <link rel="stylesheet" type="text/css" href="stylesheet.css"> 10 <link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22>ð</text></svg>"> 11 <style> 12 .projects-box { 13 /* background-color: green; */ 14 height: 160px; 15 position: relative; 16 overflow: hidden; 17 transition: all ease-in .1s; 18 19 } 20 .projects-show { 21 padding: 0; 22 margin: 0; 23 position: absolute; 24 bottom: 0; 25 left: 0; 26 text-align: center; 27 height: 30px; 28 line-height: 20px; 29 width: 100%; 30 font-size: 12px; 31 color: #0067c8; 32 cursor: pointer; 33 background: linear-gradient(to bottom, transparent, #fff, #fff); 34 } 35 .projects-show-text { 36 position: relative; 37 top: 10px; 38 } 39 </style> 40</head> 41 42<body> 43 <table style="width:100%;max-width:800px;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 44 <tr style="padding:0px"> 45 <td style="padding:0px"> 46 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 47 <tr style="padding:0px"> 48 <td style="padding:2.5%;width:63%;vertical-align:middle"> 49 <p style="text-align:center"> 50 <name>Yunfan Ye</name> 51 </p> 52 53 <p style="text-align:justify">Yunfan Ye (å¶äºå¸) is an Assistant Professor in <a href="http://design.hnu.edu.cn/gy.htm">School of Design</a>, <a href="https://www.hnu.edu.cn/">Hunan University (HNU)</a>, China. 54 I earned my Ph.D. degree in December 2023 in <a href="https://www.nudt.edu.cn/">National University of Defense Technology</a>, under the supervision of <a href="http://individual.utoronto.ca/zcai/">Prof. Zhiping Cai</a> and <a href="https://kevinkaixu.net/">Prof. Kai Xu</a> in <a href="https://kevinkaixu.net/group.html">iGrape Lab</a>. 55 I got my Master's degree in Computer Science in 2019 from <a href="https://www.stevens.edu/">Stevens Institute of Technology</a>, and Bachelor's degree in Computer Science in 2017 from <a href="https://www.xmu.edu.cn/">Xiamen University</a>, China. 56 57 </p> 58 59 <p style="text-align:center"> 60 <a href="mailto:[email protected]">Email</a>  /  61<!-- <a href="data/Yunfan-CV.pdf">CV</a>  / --> 62<!-- <a href="data/Yunfan-bio.txt">Bio</a>  / --> 63 <a href="https://scholar.google.com/citations?user=iTGg6eQAAAAJ">Google Scholar</a>  /  64 <a href="https://github.com/yunfan1202/">Github</a> 65 66 </p> 67 </td> 68 <td style="padding:2.5%;width:40%;max-width:40%"> 69 <a href="images/yeyunfan-photo.jpg"><img style="width:65%;max-width:65%" alt="profile photo" src="images/yeyunfan-photo.jpg" class="hoverZoomLink"></a> 70 </td> 71 </tr> 72 </tbody></table> 73 74 <!-- 75 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 76 <tr> 77 <td style="padding:20px;width:100%;vertical-align:middle"> 78 <heading>News</heading> 79 <ul class="projects-box" id="projects-box"> 80 <li><b>[<font color="red">2024.01</font>]</b> One paper has been accepted by <em>IEEE TIP</em>.</li> 81 82 <p class="projects-show" id="projects-show"><span class="projects-show-text" id="projects-show-text">More</span></p> 83 84 </ul> 85 </td> 86 </tr> 87 </tbody></table> 88 --> 89 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 90 <tr> 91 <td style="padding:20px;width:100%;vertical-align:middle"> 92 <heading>Teaching</heading> 93 <p> 94 æºè½è®¾è®¡æ¹æ³ (Intelligent Design Method) <a href="https://github.com/yunfan1202/intellegent_design">[Code]</a> 95 </p> 96 <p> 97 æ°æ®å¯è§å (Data Visualization) <a href="https://yunfan1202.github.io/data_visualization/fall2025/">[Project Page]</a> 98 </p> 99 </td> 100 </tr> 101 </tbody></table> 102 103 104 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 105 <tr> 106 <td style="padding:20px;width:100%;vertical-align:middle"> 107 <heading>Research</heading> 108 <p> 109 My research interest mainly include computer vision and graphics, intelligent design and their applications, especially edge detection, neural radiance field. The representative papers are <span class="highlight">highlighted</span>. 110 </p> 111 </td> 112 </tr> 113 </tbody></table> 114 115 116<!---------------------------------------------HumanSAM : Classifying Human-centric Forgery Videos in Human Spatial, Appearance, and Motion Anomaly------------------------------------------------------------------------------> 117 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 118 119<!-- --> 120 <td style="padding:20px;width:25%;vertical-align:middle"> 121 <div class="one"> 122 <img src='images/HumanSAM.jpg' height= 70%> 123 </div> 124 </td> 125 <td style="padding:20px;width:75%;vertical-align:middle"> 126 <a href="https://dejian-lc.github.io/humansam/"> 127 <papertitle>HumanSAM : Classifying Human-centric Forgery Videos in Human Spatial, Appearance, and Motion Anomaly</papertitle> 128 </a> 129 <br> 130 Chang Liu*, 131 <strong>Yunfan Ye*</strong>, 132 Fan Zhang, 133 Qingyang Zhou,
134 Yuchuan Luo, 135 <a href="http://individual.utoronto.ca/zcai/">Zhiping Cai</a> 136 <br> 137 <em>ICCV</em>, 2025 138 <br> 139 <a href="https://dejian-lc.github.io/humansam/">[Project Page]</a> 140 <a href="https://arxiv.org/abs/2507.19924">[Paper]</a> 141 <!-- 142 <a href="https://github.com/GuHuangAI/DiffusionEdge">[Code]</a> 143 --> 144 <p></p> 145 <p style="text-align:justify"> 146 Numerous synthesized videos from generative models, especially human-centric ones that simulate 147 realistic human actions, pose significant threats to human information security and authenticity. 148 While progress has been made in binary forgery video detection, the lack of fine-grained understanding 149 of forgery types raises concerns... 150 151 </p> 152 </td> 153 </tr> 154 155<!---------------------------------------------ALLVB: All-in-One Long Video Understanding Benchmark------------------------------------------------------------------------------> 156 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 157 158<!-- --> 159 <td style="padding:20px;width:25%;vertical-align:middle"> 160 <div class="one"> 161 <img src='images/ALLVB.jpg' height= 90%> 162 </div> 163 </td> 164 <td style="padding:20px;width:75%;vertical-align:middle"> 165 <a href="https://huggingface.co/datasets/ALLVB/ALLVB"> 166 <papertitle>ALLVB: All-in-One Long Video Understanding Benchmark</papertitle> 167 </a> 168 <br> 169 Xichen Tan, 170 Yuanjing Luo, 171 <strong>Yunfan Ye</strong>, 172 <a href="http://grzy.hnu.edu.cn/site/index/liufang">Fang Liu</a>, 173 <a href="http://individual.utoronto.ca/zcai/">Zhiping Cai</a> 174 <br> 175 <em>AAAI</em>, 2025 176 <br> 177 <a href="https://arxiv.org/pdf/2503.07298v2">[Paper]</a> 178 <!-- 179 <a href="https://github.com/GuHuangAI/DiffusionEdge">[Code]</a> 180 --> 181 <a href="https://huggingface.co/datasets/ALLVB/ALLVB">[Dataset]</a> 182 <p></p> 183 <p style="text-align:justify"> 184 From image to video understanding, the capabilities of Multimodal 185LLMs (MLLMs) are increasingly powerful. However, 186most existing video understanding benchmarks are relatively 187short, which makes them inadequate for effectively evaluating 188the long-sequence modeling capabilities of MLLMs... 189 190 </p> 191 </td> 192 </tr> 193 194<!---------------------------------------------Spatiotemporal-aware Neural Fields for Dynamic CT Reconstruction------------------------------------------------------------------------------> 195 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 196 197<!-- --> 198 <td style="padding:20px;width:25%;vertical-align:middle"> 199 <div class="one"> 200 <img src='images/dynamic4DCT.jpg' height= 110%> 201 </div> 202 </td> 203 <td style="padding:20px;width:75%;vertical-align:middle"> 204 <a href="https://ojs.aaai.org/index.php/AAAI/article/view/33177"> 205 <a href="https://qingyangzhou69.github.io/STNF4D/"> 206 <papertitle>Spatiotemporal-aware Neural Fields for Dynamic CT Reconstruction</papertitle> 207 </a> 208 <br> 209 Qingyang Zhou, 210 <strong>Yunfan Yeâ </strong>, 211 <a href="http://individual.utoronto.ca/zcai/">Zhiping Cai</a> 212 <br> 213 <em>AAAI</em>, 2025 214 <br> 215 <a href="https://ojs.aaai.org/index.php/AAAI/article/view/33177">[Paper]</a> 216 <!-- 217 <a href="https://arxiv.org/pdf/2401.05975.pdf">[Paper]</a> 218 <a href="https://github.com/GuHuangAI/DiffusionEdge">[Code]</a> 219 --> 220 <a href="https://qingyangzhou69.github.io/STNF4D/">[Project Page]</a> 221 <p></p> 222 <p style="text-align:justify"> 223 We propose a dynamic Computed Tomography (CT) reconstruction framework called STNF4D (SpatioTemporal-aware Neural Fields). First, we represent the 4D scene using four orthogonal volumes and compress these volumes into more compact hash grids. Compared to the plane decomposition 224method, this method enhances the modelâs capacity while
225keeping the representation compact and efficient... 226 </p> 227 </td> 228 </tr> 229 230 231 232<!---------------------------------------------DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection------------------------------------------------------------------------------> 233 <table style="width:100%;border:0px;border-spacing:0px;border-collapse:separate;margin-right:auto;margin-left:auto;"><tbody> 234 <tr bgcolor="#ffffd0"> 235<!-- --> 236 <td style="padding:20px;width:25%;vertical-align:middle"> 237 <div class="one"> 238 <img src='images/diffusion_edge.png' height= 90%> 239 </div> 240 </td> 241 <td style="padding:20px;width:75%;vertical-align:middle"> 242 <a href="https://arxiv.org/pdf/2401.02032.pdf"> 243 <papertitle>DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection</papertitle> 244 </a> 245 <br> 246 <strong>Yunfan Ye*</strong>, 247 <a href="https://kevinkaixu.net/">Kai Xu*</a>, 248 <a href="https://github.com/GuHuangAI">Yuhang Huangâ </a>, 249 <a href="https://renjiaoyi.github.io/">Renjiao Yi</a>, 250 <a href="http://individual.utoronto.ca/zcai/">Zhiping Cai</a> 251 <br> 252 <em>AAAI</em>, 2024 253 <br> 254 <a href="https://ojs.aaai.org/index.php/AAAI/article/view/28490">[Paper]</a> 255 <a href="https://github.com/GuHuangAI/DiffusionEdge">[Code]</a> 256 <a href="https://mp.weixin.qq.com/s/rHMO1nFCvz44tLVcpDWOPw">[News]</a> 257 <p></p> 258 <p style="text-align:justify"> 259 Limited by the encoder-decoder architecture, learning-based edge detectors usually have difficulty predicting edge maps that satisfy both correctness and crispness. With the recent success of the diffusion probabilistic model (DPM), we found it is especially suitable for accurate and crisp edge detection... 260 </p> 261 </td> 262 </tr> 263<!---------------------------------------------NEF: Neural Edge Fields for 3D Parametric Curve Reconstruction from Multi-view Images------------------------------------------------------------------------------> 264 <tr bgcolor="#ffffd0"> 265<!-- --> 266 <td style="padding:20px;width:25%;vertical-align:middle"> 267 <div class="one"> 268 <img src='images/NEF.png' width=112%> 269 </div> 270 </td> 271 <td style="padding:20px;width:75%;vertical-align:middle"> 272 <a href="https://openaccess.thecvf.com/content/CVPR2023/papers/Ye_NEF_Neural_Edge_Fields_for_3D_Parametric_Curve_Reconstruction_From_CVPR_2023_paper.pdf"> 273 <papertitle>NEF: Neural Edge Fields for 3D Parametric Curve Reconstruction from Multi-view Images</papertitle> 274 </a> 275 <br> 276 <strong>Yunfan Ye</strong>, 277 <a href="https://renjiaoyi.github.io/">Renjiao Yi</a>, 278 <a href="https://github.com/zhirui-gao">Zhirui Gao</a>, 279 <a href="https://www.zhuchenyang.net/">Chenyang Zhu</a>, 280 <a href="http://individual.utoronto.ca/zcai/">Zhiping Cai</a>, 281 <a href="https://kevinkaixu.net/">Kai Xuâ </a> 282 <br> 283 <em>CVPR</em>, 2023 284 <br> 285 <a href="https://openaccess.thecvf.com/content/CVPR2023/papers/Ye_NEF_Neural_Edge_Fields_for_3D_Parametric_Curve_Reconstruction_From_CVPR_2023_paper.pdf">[Paper]</a> 286 <a href="https://github.com/yunfan1202/NEF_code">[Code]</a> 287 <a href="https://yunfan1202.github.io/NEF/">[Project Page]</a> 288 <p></p> 289 <p style="text-align:justify"> 290 We study the problem of reconstructing 3D feature curves of an object from a set of calibrated multi-view images. To do so, we learn a neural implicit field representing the density distribution of 3D edges which we refer to as Neural Edge Field (NEF). Inspired by NeRF... 291 </p> 292 </td> 293 </tr> 294<!------------------------------------------------Delving into Crispness: Guided Label Refinement for Crisp Edge Detection---------------------------------------------------------------------------> 295 <tr> 296<!-- --> 297 <td style="padding:20px;width:25%;vertical-align:middle"> 298 <div class="one"> 299 <img src='images/crisp_edge.png' width=115%> 300 </div> 301 </td> 302 <td style="padding:20px;width:75%;vertical-align:middle"> 303 <a href="https://arxiv.org/pdf/2306.15172.pdf"> 304 <papertitle>Delving into Crispness: Guided Label Refinement for Crisp Edge Detection</papertitle> 305 </a> 306 <br> 307 <strong>Yunfan Ye</strong>, 308 <a href="https://renjiaoyi.github.io/">Renjiao Yi</a>, 309 <a href="https://github.com/zhirui-gao">Zhirui Gao</a>, 310 <a href="http://individual.utoronto.ca/zcai/">Zhiping Caiâ </a>, 311 <a href="https://kevinkaixu.net/">Kai Xuâ </a> 312 <br> 313 <em>IEEE TIP</em>, 2023 314 <br> 315 <a href="https://arxiv.org/pdf/2306.15172.pdf">[Paper]</a> 316 <a href="https://github.com/yunfan1202/Delving-into-Crispness">[Code]</a> 317 318 <p></p> 319 <p style="text-align:justify"> 320 Learning-based edge detection usually suffers from predicting thick edges. Through extensive quantitative study with a new edge crispness measure, we find that noisy human-labeled edges are the main cause of thick predictions... 321 322 </p> 323 </td> 324 </tr> 325<!---------------------------------------------------------------------------------------------------------------------------> 326<!---------------------------------------------------------------------------------------------------------------------------> 327 <tr> 328 <td style="padding:20px;width:25%;vertical-align:middle"> 329 <div class="one"> 330 <img src='images/STEdge.png' width=115%> 331 </div> 332 </td> 333 <td style="padding:20px;width:75%;vertical-align:middle"> 334 <a href="https://arxiv.org/pdf/2201.05121.pdf"> 335 <papertitle>STEdge: Self-Training Edge Detection With Multilayer Teaching and Regularization</papertitle> 336 </a> 337 <br> 338 <strong>Yunfan Ye*</strong>, 339 <a href="https://renjiaoyi.github.io/">Renjiao Yi*</a>, 340 <a href="http://individual.utoronto.ca/zcai/">Zhiping Caiâ </a>, 341 <a href="https://kevinkaixu.net/">Kai Xuâ </a> 342 <br> 343 <em>IEEE TNNLS</em>, 2023 344 <br> 345 <a href="https://arxiv.org/pdf/2201.05121.pdf">[Paper]</a> 346 <a href="https://github.com/yunfan1202/STEdge">[Code]</a> 347 <p></p> 348 <p style="text-align:justify"> 349 Learning-based edge detection has hereunto been strongly supervised with pixel-wise annotations which are tedious to obtain manually. We study the problem of self-training edge detection, leveraging the untapped wealth of large-scale unlabeled image datasets... 350 </p> 351 </td> 352 </tr> 353 354<!---------------------------------------------------------------------------------------------------------------------------> 355<!---------------------------------------------------------------------------------------------------------------------------> 356 <tr> 357 <td style="padding:20px;width:25%;vertical-align:middle"> 358 <div class="one"> 359 <img src='images/watermarking.jpg' width=115%> 360 </div> 361 </td> 362 <td style="padding:20px;width:75%;vertical-align:middle"> 363 <a href="https://ieeexplore.ieee.org/document/10184464"> 364 <papertitle>Fixing the Double Agent Vulnerability of Deep Watermarking: A Patch-Level Solution against Artwork Plagiarism</papertitle> 365 </a> 366 <br> 367 <a href="https://github.com/1024yy">Yuanjing Luo*</a>, 368 <a href="https://tongqingzhou-nudt.github.io/">
368Tongqing Zhou*</a>, 369 Shenglan Cui, 370 <strong>Yunfan Ye</strong>, 371 <a href="http://design.hnu.edu.cn/info/1023/5787.htm">Fang Liuâ </a>, 372 Zhiping Cai 373 <br> 374 <em>IEEE TCSVT</em>, 2023 375 <br> 376 <a href="https://ieeexplore.ieee.org/document/10184464">[Paper]</a> 377 <a href="https://github.com/1024yy/DIPW">[Code]</a> 378 <p></p> 379 <p style="text-align:justify"> 380Increasing artwork plagiarism incidents stresses the urgent need for proper copyright protection on behalf of the creators. The latest development in this context focuses on embedding watermarks via deep encoder-decoder networks... 381 </td> 382 </tr> 383<!---------------------------------------------------------------------------------------------------------------------------> 384<!---------------------------------------------------------------------------------------------------------------------------><!---------------------------------------------------------------------------------------------------------------------------> 385 <tr> 386 <td style="padding:20px;width:25%;vertical-align:middle"> 387 <div class="one"> 388 <img src='images/template.jpg' width=115%> 389 </div> 390 </td> 391 <td style="padding:20px;width:75%;vertical-align:middle"> 392 <a href="https://arxiv.org/pdf/2303.08438.pdf"> 393 <papertitle>Learning Accurate Template Matching with Differentiable Coarse-to-Fine Correspondence Refinement</papertitle> 394 </a> 395 <br> 396 <a href="https://github.com/zhirui-gao">Zhirui Gao</a>, 397 <a href="https://renjiaoyi.github.io/">Renjiao Yi</a>, 398 <a href="https://github.com/qinzheng93">Zheng Qin</a>, 399 <strong>Yunfan Ye</strong>, 400 <a href="https://www.zhuchenyang.net/">Chenyang Zhu</a>, 401 <a href="https://kevinkaixu.net/">Kai Xuâ </a> 402 <br> 403 <br> 404 <em>Computational Visual Media Journal (CVMJ)</em> 405 <br> 406 <a href="https://arxiv.org/pdf/2303.08438.pdf">[Paper]</a> 407 <a href="https://github.com/zhirui-gao/Deep-Template-Matching">[Code]</a> 408 <p></p> 409 <p style="text-align:justify"> 410Template matching is a fundamental task in computer vision and has been studied for decades. It plays an essential role in manufacturing industry... 411 </td> 412 </tr> 413 414<!---------------------------------------------------------------------------------------------------------------------------> 415 416<!---------------------------------------------------------------------------------------------------------------------------><!---------------------------------------------------------------------------------------------------------------------------> 417 <tr> 418 <td style="padding:20px;width:25%;vertical-align:middle"> 419 <div class="one"> 420 <img src='images/caption.jpg' width=115%> 421 </div> 422 </td> 423 <td style="padding:20px;width:75%;vertical-align:middle"> 424 <a href="https://link.springer.com/article/10.1007/s00530-023-01178-8"> 425 <papertitle>Image Captioning for Cultural Artworks: a Case Study on Ceramics</papertitle> 426 </a> 427 <br> 428 Baoying Zheng, 429 <a href="http://design.hnu.edu.cn/info/1023/5787.htm">Fang Liuâ </a>, 430 Mohan Zhang, 431 <a href="https://tongqingzhou-nudt.github.io/">Tongqing Zhou</a>, 432 Shenglan Cui, 433 <strong>Yunfan Ye</strong>, 434 Yeting Guo 435 <br> 436 <br> 437 <em>Multimedia System</em> 438 <br> 439 <a href="https://link.springer.com/article/10.1007/s00530-023-01178-8">[Paper]</a> 440 <p></p> 441 <p style="text-align:justify"> 442 When viewing ancient artworks, people try to build connections with them to âreadâ the correct messages from the past. A proper descriptive caption is essential for viewers... 443 </td> 444 </tr> 445<!--------------------------------------------------------------------------------------------------------------------------->
446 <table width="100%" align="center" border="0" cellspacing="0" cellpadding="20"> 447 </table> 448 449 <p align="center">Source code from <a href="https://jonbarron.info/">Jon Barron</a>'s website</p> 450
450<script> 451 let show = true; 452 document.querySelector('#projects-show').onclick = function() { 453 if (!show) { 454 document.querySelector('#projects-box').style.height = 'auto'; 455 document.querySelector('#projects-box').style.paddingBottom = '20px'; 456 document.querySelector('#projects-show-text').innerHTML = 'Less'; 457 show = true; 458 } else { 459 show = false; 460 document.querySelector('#projects-box').style.height = '160px'; 461 document.querySelector('#projects-box').style.paddingBottom = '0px'; 462 document.querySelector('#projects-show-text').innerHTML = 'More'; 463 } 464 } 465 </script>
465 466</body> 467</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.