1<!DOCTYPE html> 2<html> 3<head> 4 <meta charset="utf-8"> 5 <meta name="description" 6 content="Clearer Frames, Anytime: Resolving Velocity Ambiguity in Video Frame Interpolation."> 7 <meta name="keywords" content="video frame interpolation, anytime, clearer, manipulated interpolation of anything"> 8 <meta name="viewport" content="width=device-width, initial-scale=1"> 9 <title>InterpAny-Clearer</title> 10 11 <!-- Global site tag (gtag.js) - Google Analytics -->
vendor: 68 bytes, lines 11-12
11 12 <script async src="https://www.googletagmanager.com/gtag/js?id=
12G-EDF010G6PN
vendor: 16 bytes, lines 12-13
12"></script> 13
13<script> 14
vendor: 177 bytes, lines 14-22
14window.dataLayer = window.dataLayer || []; 15 16 function gtag() { 17 dataLayer.push(arguments); 18 } 19 20 gtag('js', new Date()); 21 22 gtag('config', '
22G-EDF010G6PN
vendor: 10 bytes, lines 22-25
22'); 23 24 25
25</script>
25 26 27
27<script type="module" 28 src="https://unpkg.com/@google/model-viewer/dist/model-viewer.min.js"></script>
28 29 30
30<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.6.0/jquery.min.js"></script>
30 31
31<script type="text/javascript" src="https://code.jquery.com/jquery-1.11.0.min.js"></script>
31 32
32<script type="text/javascript" src="https://code.jquery.com/jquery-migrate-1.2.1.min.js"></script>
32 33
33<script src="https://unpkg.com/interactjs/dist/interact.min.js"></script>
33 34
34<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
34 35 36 <link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" 37 rel="stylesheet"> 38 39 <link rel="stylesheet" type="text/css" href="./static/slick/slick.css"/> 40 <link rel="stylesheet" type="text/css" href="./static/slick/slick-theme.css"/> 41 42 <link rel="stylesheet" href="./static/css/bulma.min.css"> 43 <link rel="stylesheet" href="./static/css/bulma-slider.min.css"> 44 <link rel="stylesheet" href="./static/css/fontawesome.all.min.css"> 45 <link rel="stylesheet" 46 href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css"> 47 <link rel="stylesheet" href="./static/css/index.css"> 48<!-- <link rel="icon" href="./static/images/logo.png">--> 49 50
50<script defer src="./static/js/fontawesome.all.min.js"></script>
50 51
51<script src="./static/js/bulma-slider.min.js"></script>
51 52
52<script src="./static/js/index.js"></script>
52 53</head> 54<body> 55 56<nav class="navbar" role="navigation" aria-label="main navigation"> 57 <div class="navbar-brand"> 58 <a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false"> 59 <span aria-hidden="true"></span> 60 <span aria-hidden="true"></span> 61 <span aria-hidden="true"></span> 62 </a> 63 </div> 64 <div class="navbar-menu"> 65 <div class="navbar-start" style="flex-grow: 1; justify-content: center;"> 66 <a class="navbar-item" href="https://zzh-tech.github.io/"> 67 <span class="icon"> 68 <i class="fas fa-home"></i> 69 </span> 70 </a> 71 72 <div class="navbar-item has-dropdown is-hoverable"> 73 <a class="navbar-link"> 74 More Research 75 </a> 76 <div class="navbar-dropdown"> 77 <a class="navbar-item" href="https://zzh-tech.github.io/BiT/"> 78 BiT 79 </a> 80 <a class="navbar-item" href="https://zzh-tech.github.io/Animation-from-Blur/"> 81 Animation-from-Blur 82 </a> 83 <a class="navbar-item" href="https://zzh-tech.github.io/Dual-Reversed-RS/"> 84 Dual-Reversed-RS 85 </a> 86 </div> 87 </div> 88 </div> 89 90 </div> 91</nav> 92 93 94<section class="hero"> 95 <div class="hero-body"> 96 <div class="container"> 97 <!-- <div class="columns is-centered">--> 98 <!-- <div class="column is-4 has-text-centered">--> 99 <!-- <img src="static/images/logo.png" alt="InterpAny-Clearer logo" width="288" style="border-radius: 8%;"/>--> 100 <!-- </div>--> 101 <!-- </div>--> 102 <br> 103 <div class="container has-text-centered"> 104 <h1 class="title is-1 publication-title"> 105 <u style="color: hotpink; text-decoration: underline dotted hotpink;">Clearer Frames,</u> 106 <u style="color: dodgerblue; text-decoration: underline dotted dodgerblue;">Anytime:</u> Resolving 107 Velocity Ambiguity in Video Frame 108 Interpolation </h1> 109 <div class="is-size-5 publication-authors"> 110 <span class="author-block"><a 111 href="https://zzh-tech.github.io/">Zhihang Zhong</a><sup>1,*</sup>, </span> 112 <span class="author-block"><a 113 href="https://scholar.google.com/citations?user=BKYVv4MAAAAJ&hl=en">Gurunandan Krishnan</a><sup>2</sup>, </span> 114 <span class="author-block"><a 115 href="https://jimmysuen.github.io/">Xiao Sun</a><sup>1</sup>, </span> 116 <span class="author-block"><a 117 href="https://scholar.google.com/citations?user=gFtI-8QAAAAJ&hl=en">Yu Qiao</a><sup>1</sup>, </span> 118 <span class="author-block"><a 119 href="https://sizhuoma.netlify.app/">Sizhuo Ma</a><sup>2,â </sup>, </span> 120 <span class="author-block"><a 121 href="https://jianwang-cmu.github.io/">Jian Wang</a><sup>2,â </sup></span> 122 </div> 123 124 <div class="is-size-5 publication-authors"> 125 <span class="author-block"><sup>1</sup><a href="https://github.com/OpenGVLab">Shanghai AI Laboratory, OpenGVLab</a>, </span> 126 <span class="author-block"><sup>2</sup><a href="https://snap.com/en-US">Snap Inc.</a></span> 127 <br> 128 <span class="author-block"><sup>*</sup>First author,</span> 129 <span class="author-block"><sup>â </sup>Co-corresponding authors</span> 130 </div> 131 132 <div class="column has-text-centered"> 133 <div class="publication-links"> 134 <!-- <span class="link-block">--> 135 <!-- <a href="https://github.com/zzh-tech/InterpAny-Clearer"--> 136 <!-- class="external-link button is-normal is-rounded is-dark">-->
137 <!-- <span class="icon">--> 138 <!-- <i class="fas fa-file-pdf"></i>--> 139 <!-- </span>--> 140 <!-- <span>Paper</span>--> 141 <!-- </a>--> 142 <!-- </span>--> 143 <span class="link-block"> 144 <a href="http://arxiv.org/abs/2311.08007" 145 class="external-link button is-normal is-rounded is-dark" 146 target="_blank"> 147 <span class="icon"> 148 <i class="ai ai-arxiv"></i> 149 </span> 150 <span>arXiv</span> 151 </a> 152 </span> 153 <span class="link-block"> 154 <a href="https://github.com/zzh-tech/InterpAny-Clearer" 155 class="external-link button is-normal is-rounded is-dark" 156 target="_blank"> 157 <span class="icon"> 158 <i class="fab fa-github"></i> 159 </span> 160 <span>Code</span> 161 </a> 162 </span> 163 <span class="link-block"> 164 <a href="http://ai4sports.opengvlab.com/interpany-clearer/" 165 class="external-link button is-normal is-rounded is-dark" 166 target="_blank"> 167 <span class="icon"> 168 <i class="fa fa-star"></i> 169 </span> 170 <span>Demo</span> 171 </a> 172 </span> 173 <div class="is-size-5 publication-authors"> 174 <img src="./static/images/eccv-navbar-logo.svg" 175 style="width: 18%;text-align: center"> 176 <br> 177 <span style="color:darkred"><b>Accepted to ECCV'2024 as Oral</b></span> 178 </div> 179 </div> 180 </div> 181 </div> 182 </div> 183 </div> 184</section> 185 186<section class="hero teaser"> 187 <div class="hero-body"> 188 <div class="container is-max-desktop"> 189 <img src="./static/images/teaser.jpg" alt="teaser"> 190 <br> 191 <br> 192 <h2 class="subtitle has-text-centered"> 193 <b> 194 We addressed velocity ambiguity in video frame interpolation, leading to: <br> 195 <em style="color: red">Clearer anytime frame interpolation</em> & <em style="color: red">Manipulated 196 interpolation of anything</em>. 197 </b> 198 </h2> 199 </div> 200 </div> 201</section> 202 203<br> 204 205<div class="columns is-centered has-text-centered"> 206 <div class="column is-four-fifths"> 207 <h2 class="title is-3">Clearer anytime frame interpolation</h2> 208 </div> 209</div> 210<br> 211 212<section class="hero is-dark is-small"> 213 <div class="hero-body"> 214 <div class="container has-text-centered"> 215 <div id="results-carousel" class="carousel results-carousel"> 216 <div class="results-item"> 217 <video poster="" id="00006/0277" autoplay controls muted loop playsinline height="100%"> 218 <source src="./static/webapp_demo/00006/0277_demo_converted.mp4" 219 type="video/mp4"> 220 </video> 221 </div> 222 223 <div class="results-item"> 224 <video poster="" id="00029/0419" autoplay controls muted loop playsinline height="100%"> 225 <source src="./static/webapp_demo/00029/0419_demo_converted.mp4" 226 type="video/mp4"> 227 </video> 228 </div> 229 230 <div class="results-item"> 231 <video poster="" id="00036/0310" autoplay controls muted loop playsinline height="100%"> 232 <source src="./static/webapp_demo/00036/0310_demo_converted.mp4" 233 type="video/mp4"> 234 </video> 235 </div> 236 237 <div class="results-item"> 238 <video poster="" id="00036/0320" autoplay controls muted loop playsinline height="100%"> 239 <source src="./static/webapp_demo/00036/0320_demo_converted.mp4" 240 type="video/mp4"> 241 </video> 242 </div> 243 244 <div class="results-item"> 245 <video poster="" id="00037/0345" autoplay controls muted loop playsinline height="100%"> 246 <source src="./static/webapp_demo/00037/0345_demo_converted.mp4" 247 type="video/mp4"> 248 </video> 249 </div> 250 251 <div class="results-item"> 252 <video poster="" id="00068/0261" autoplay controls muted loop playsinline height="100%"> 253 <source src="./static/webapp_demo/00068/0261_demo_converted.mp4" 254 type="video/mp4"> 255 </video> 256 </div> 257 258 <div class="results-item"> 259 <video poster="" id="00071/0366" autoplay controls muted loop playsinline height="100%"> 260 <source src="./static/webapp_demo/00071/0366_demo_converted.mp4" 261 type="video/mp4"> 262 </video> 263 </div> 264 265 <div class="results-item"> 266 <video poster="" id="00071/0808" autoplay controls muted loop playsinline height="100%"> 267 <source src="./static/webapp_demo/00071/0808_demo_converted.mp4" 268 type="video/mp4"> 269 </video> 270 </div> 271 272 <div class="results-item"> 273 <video poster="" id="00080/0050" autoplay controls muted loop playsinline height="100%"> 274 <source src="./static/webapp_demo/00080/0050_demo_converted.mp4" 275 type="video/mp4"> 276 </video> 277 </div> 278 279 </div> 280 281 <div class="columns is-centered has-text-centered"> 282 <div class="column is-three-quarters"> 283 <p> 284 <em> 285 <b>When integrating our plug-and-play training strategies ([D,R]) into the state-of-the-art 286 learning-based models such as 287 <a href="https://github.com/megvii-research/ECCV2022-RIFE">RIFE [1]</a>, 288 <a href="https://github.com/ltkong218/IFRNet">IFRNet [2]</a>, 289 <a href="https://github.com/MCG-NKU/AMT">AMT [3]</a>, and 290 <a href="https://github.com/ltkong218/IFRNet">EMA-VFI [4]</a>, they exhibit markedly 291 sharper outputs and superior perceptual quality in arbitrary time interpolations. <br> 292 (Here, we employ RIFE as an illustrative example, generating 128 interpolated frames
293 using just two images.) 294 </b> 295 </em> 296 </p> 297 </div> 298 </div> 299 </div> 300 </div> 301</section> 302 303<br> 304<br> 305 306<section class="section"> 307 <!-- Paper video. --> 308 <div style="display: flex; justify-content: center; align-items: center; height: 75vh; width: 100%;"> 309 <div style="width: 66%; height: auto;"> 310 <div class="columns is-centered has-text-centered"> 311 <div class="column is-four-fifths"> 312 <h2 class="title is-3">Application in CCTV</h2> 313 <br> 314 <div class="publication-video"> 315 <iframe src="./static/webapp_demo/cctv5_interpany-clearer.mp4" encrypted-media allowfullscreen></iframe> 316 </div> 317 <br> 318 <div class="columns is-centered has-text-centered"> 319 <div class="column is-three-quarters"> 320 <p> 321 <em> 322 <b> 323 <u style="color: black; text-decoration: underline black;"> 324 Our technology was used by CCTV5 as well as CCTV5+ for slow motion 325 demonstrations of athletes jumping in the 2024 Thomas & Uber Cup.</u> 326 </b> 327 </em> 328 </p> 329 </div> 330 </div> 331 </div> 332 </div> 333 </div> 334 </div> 335 <!--/ Paper video. --> 336</section> 337 338<br> 339<br> 340 341<section class="section"> 342 <!-- Paper video. --> 343 <div style="display: flex; justify-content: center; align-items: center; height: 75vh; width: 100%;"> 344 <div style="width: 66%; height: auto;"> 345 <div class="columns is-centered has-text-centered"> 346 <div class="column is-four-fifths"> 347 <h2 class="title is-3">Manipulated interpolation of anything</h2> 348 <br> 349 <div class="publication-video"> 350 <iframe src="./static/webapp_demo/InterpAny-Clearer-Demo.mp4" 351 allow="autoplay; encrypted-media" allowfullscreen></iframe> 352 </div> 353 <br> 354 <div class="columns is-centered has-text-centered"> 355 <div class="column is-three-quarters"> 356 <p> 357 <em> 358 <b> 359 Additionally, our strategies enable temporal manipulation of each object 360 independently during the inference stage, offering a novel tool for video 361 editing tasks like re-timing. <u 362 style="color: black; text-decoration: underline black;">Our <a 363 href="http://ai4sports.opengvlab.com/interpany-clearer/" 364 target="_blank">APP</a> also 365 supports video as an input to classic video frame interpolation.</u> 366 </b> 367 </em> 368 </p> 369 </div> 370 </div> 371 </div> 372 </div> 373 </div> 374 </div> 375 <!--/ Paper video. --> 376</section> 377 378<hr/> 379 380<section class="section"> 381 <div class="container is-max-desktop"> 382 <!-- Abstract. --> 383 <div class="columns is-centered has-text-centered"> 384 <div class="column is-four-fifths"> 385 <h2 class="title is-3">Abstract</h2> 386 <div class="content has-text-justified"> 387 <p> 388 Existing video frame interpolation (VFI) methods blindly predict where each object is at a 389 specific timestep t ("time indexing"), which struggles to predict precise object movements. 390 Given two images of a ⚾, there are infinitely many possible trajectories: accelerating or 391 decelerating, straight or curved. This often results in blurry frames as the method averages out 392 these possibilities. Instead of forcing the network to learn this complicated time-to-location 393 mapping implicitly together with predicting the frames, we provide the network with an explicit 394 hint on how far the object has traveled between start and end frames, a novel approach termed 395 "distance indexing". This method offers a clearer learning goal for models, reducing the 396 uncertainty tied to object speeds. We further observed that, even with this extra guid
396ance, 397 objects can still be blurry especially when they are equally far from both input frames 398 (<i>i.e.</i>, halfway in-between), due to the directional ambiguity in long-range motion. To 399 solve this, we propose an iterative reference-based estimation strategy that breaks down a 400 long-range prediction into several short-range steps. When integrating our plug-and-play 401 strategies into state-of-the-art learning-based models, they exhibit markedly sharper outputs 402 and superior perceptual quality in arbitrary time interpolations, using a uniform distance 403 indexing map in the same format as time indexing. Additionally, distance indexing can be 404 specified pixel-wise, which enables temporal manipulation of each object independently, offering 405 a novel tool for video editing tasks like re-timing. 406 </p> 407 </div> 408 </div> 409 </div> 410 <!--/ Abstract. --> 411 412 <!-- <!– Paper video. –>--> 413 <!-- <div class="columns is-centered has-text-centered">--> 414 <!-- <div class="column is-four-fifths">--> 415 <!-- <h2 class="title is-2">Video</h2>--> 416 <!-- <div class="publication-video">--> 417 <!-- <iframe width="640" height="480" src="https://www.youtube.com/embed/qzgdE_ghkaI"--> 418 <!-- title="YouTube video player" frameborder="0"--> 419 <!-- allow="accelerometer; clipboard-write; encrypted-media; gyroscope; picture-in-picture"--> 420 <!-- allowfullscreen></iframe>--> 421 <!-- </div>--> 422 <!-- </div>--> 423 <!-- </div>--> 424 <!-- <!–/ Paper video. –>--> 425 </div> 426</section> 427 428<section> 429 <div class="container is-max-desktop"> 430 <div class="content has-text-justified"> 431 <h3 class="title is-4">Velocity ambiguity problem</h3> 432 <p> 433 The current mainstream algorithm for arbitrary-time frame interpolation, denoted as \(\mathcal{F}\), 434 predicts the target frame \(I_t\) using the starting frame \(I_0\), the ending frame \(I_1\), and a time 435 index 436 \(t\) as inputs: 437 438 $$I_t = \mathcal{F}\left(I_0, I_1, t\right)$$ 439 440 However, the unknown motion velocity of each independently moving object introduces the issue of 441 <b>"velocity ambiguity"</b>. This means there are multiple possible mappings from the same inputs to 442 different locations: 443 444 $$\left\{I_t^1, I_t^2, \ldots, I_t^n\right\} = \mathcal{F}(I_0, I_1, t)$$ 445 446 Taking a ⚾ as an example, there are countless potential landing spots for the ball in mid-air, 447 leading to conflicts in learning during the training process. 448 449 In short, the algorithm cannot discern which scenario is correct to learn, so it settles for an average 450 state: 451 452 $$ \hat{I}_t = \mathbb{E}_{I_t \sim \mathcal{F}(I_0, I_1, t)}[I_t] $$ 453 454 This results in the algorithm's prediction \(\hat{I}_t\) being blurry during testing: 455 </p> 456 <div class="has-text-centered"> 457 <img style="width: 100%;" src="./static/images/velocity_ambiguity.jpg" 458 alt="Velocity ambiguity"/> 459 </div> 460 </div> 461 </div> 462</section> 463 464<br> 465<br> 466 467<section> 468 <div class="container is-max-desktop"> 469 <div class="content has-text-justified"> 470 <h3 class="title is-4">Disambiguation</h3> 471 <p> 472 To resolve velocity ambiguity, a new paradigm for index-based learning is required. 473 <i><b>We need to guide the algorithm on why objects land in certain positions:</b></i> 474 $$I_{t} = \mathcal{F}\left(I_0, I_1, \text{motion hint}\right)$$ 475 </p> 476 <h4 class="title is-5">Distance indexing</h4> 477 <p> 478 In fact, the training approach using time indexing requires the algorithm to not only learn how to 479 interpolate frames but also to guess the mapping relationship from time to position, denoted as 480 \(\mathcal{D}\): 481 482 $$I_t = \mathcal{F}(I_0, I_1, t) \to I_t = \mathcal{F}(I_0, I_1, \mathcal{D}(t))$$ 483 484 Our solution involves calculating a path distance ratio map \(D_t\) to replace the time \(t\) for 485 index-based learning: 486 487 $$I_t = \mathcal{F}(I_0, I_1, \mathcal{D}(t)) \to I_t = \mathcal{F}(I_0, I_1, D_t)$$ 488 </p> 489 <div class="has-text-centered"> 490 <img style="width: 100%;" src="./static/images/distance_indexing.jpg" 491 alt="Distance indexing"/> 492 </div> 493 <br> 494 <p> 495 We first compute the optical flows from \(I_0\) to \(I_t\) and from \(I_0\) to \(I_1\), denoted as 496 \(V_{0\to t}\) and \(V_{0\to 1}\), respectively. Then, for each pixel location \((x,y)\), we calculate 497 the proportion of the optical flow \(V_{0\to t}\) projected onto \(V_{0\to 1}\), denoted as the "path 498 distance ratio": 499 500 $$D_t(x,y) = \frac{\left\Vert \mathbf{V}_{0\to t}(x,y)\right\Vert \cos{\theta}}{\left\Vert 501 \mathbf{V}_{0\to 1}(x,y) \right\Vert}$$ 502 503 <i><b>With \(D_t\), the algorithm avoids the ambiguous time-to-position mappings during training caused 504 by varying velocities, allowing for clearer predictions during testing. Importantly, even without 505 the ability to calculate the exact \(D_t\) using ground truth labels during inference, providing a 506 uniform index map similar to time indexing, <i>i.e.</i>, \(D_t=t\), the algorithm can still predict 507 clearer images (simulating uniform motion of objects).</b></i> 508 </p> 509 <br> 510 <h4 class="title is-5">Iterative reference-based estimation</h4> 511 <p> 512 While path distance indexing helps us sidestep ambiguities in speed, it does not resolve directional 513 ambiguities. We apply a classic divide-and-conquer strategy to minimize the impact of directional 514 ambiguities, further improving the predicted outcome. 515 <br> 516 <b>In practice, we break down a long-distance inference into a series of short-distance inferences from 517 near to far, using the previous inference, along with the starting and ending frames, as references 518 to avoid accumulated errors:</b> 519 520 $$I_t = \mathcal{F}(I_0, I_1, D_t, I_{\text{ref}}, D_{\text{ref}})$$ 521 522 For example, dividing the inference into two steps would look like: 523 524 $$I_{t/2} = \mathcal{F}(I_0, I_1, D_{t/2}, I_{0}, D_{0})$$ 525 526 $$I_{t} = \mathcal{F}(I_0, I_1, D_{t}, I_{t/2}, D_{t/2})$$ 527 528 Similarly, taking a ⚾ as an example, our proposed strategies are illustrated in the following 529 figure: 530 </p> 531 <div class="has-text-centered"> 532 <img style="width: 100%;" src="./static/images/disambiguation.jpg" 533 alt="Disambiguation"/> 534 </div> 535 <br> 536 <br> 537 <h4 class="title is-5">Editable interpolation</h4> 538 <p> 539 <b>Beyond using a uniform index map like time indexing, we can also take advantage of the 2D editable 540 nature of path distance indexing to implement editable frame interpolation techniques.</b> 541 Initially, we can obtain masks for objects of interest using the Segment Anything Model (SAM) [5]. We 542 then 543 customize the path distance curves for different object regions to achieve manipulated interpolation of 544 anything. 545 </p> 546 <div class="has-text-centered"> 547 <img style="width: 100%;" src="./static/images/manipulation.jpg" 548 alt="Manipulation"/> 549 </div> 550 </div> 551 </div> 552</section> 553 554<br> 555<br> 556 557<section> 558 <div class="container is-max-desktop"> 559 <div class="content has-text-justified"> 560 <h2 class="title is-3">Ending</h2> 561 <p> 562 This work presents the next generation of video frame interpolation technology, aiming to inspire 563 readers and contribute to fields like video enhancement, editing, and generation! ð¥ð¥ð¥ 564 <br> 565 Welcome to 🌟 <a href="https://github.com/zzh-tech/InterpAny-Clearer">this project</a> 566 and follow <a href="https://github.com/zzh-tech">the author's GitHub</a>ï½ 567 </p> 568 </div> 569 </div> 570</section> 571 572<br> 573<hr/> 574 575<section class="section" id="BibTeX"> 576 <div class="container is-max-desktop content"> 577 <h2 class="title">BibTeX</h2> 578 <pre><code>@article{zhong2023clearer, 579 title={Clearer Frames, Anytime: Resolving Velocity Ambiguity in Video Frame Interpolation}, 580 author={Zhong, Zhihang and Krishnan, Gurunandan and Sun, Xiao and Qiao, Yu and Ma, Sizhuo and Wang, Jian}, 581 journal={arXiv preprint arXiv:2311.08007}, 582 year={2023}}</code></pre> 583 </div> 584</section> 585 586<section class="section" id="acknowledgements"> 587 <div class="container content is-max-desktop"> 588 <h2 class="title">Acknowledgements</h2> 589 <p> 590 We thank Dorian Chan, Zhirong Wu, and Stephen Lin for their insightful feedback and advice. Our thanks also 591 go to Vu An Tran for developing the web application, and to Wei Wang for coordinating the user study. 592 </p> 593 </div> 594</section> 595 596<section class="section" id="reference"> 597 <div class="container content is-max-desktop"> 598 <h2 class="title">Reference</h2> 599 <p> 600 [1] Huang, Zhewei, Tianyuan Zhang, Wen Heng, Boxin Shi, and Shuchang Zhou. "Real-time intermediate flow 601 estimation for video frame interpolation." In European Conference on Computer Vision, pp. 624-642. Cham: 602 Springer Nature Switzerland, 2022. 603 <br> 604 [2] Kong, Lingtong, Boyuan Jiang, Donghao Luo, Wenqing Chu, Xiaoming Huang, Ying Tai, Chengjie Wang, and Jie 605 Yang. "Ifrnet: Intermediate feature refine network for efficient frame interpolation." In Proceedings of the 606 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1969-1978. 2022. 607 <br> 608 [3] Li, Zhen, Zuo-Liang Zhu, Ling-Hao Han, Qibin Hou, Chun-Le Guo, and Ming-Ming Cheng. "AMT: All-Pairs 609 Multi-Field Transforms for Efficient Frame Interpolation." In Proceedings of the IEEE/CVF Conference on 610 Computer Vision and Pattern Recognition, pp. 9801-9810. 2023. 611 <br> 612 [4] Zhang, Guozhen, Yuhan Zhu, Haonan Wang, Youxin Chen, Gangshan Wu, and Limin Wang. "Extracting motion and 613 appearance via inter-frame attention for efficient video frame interpolation." In Proceedings of the 614 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 5682-5692. 2023. 615 <br> 616 [5] Kirillov, Alexander, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, 617 Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollár, and Ross Girshick. "Segment Anything." In 618 Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 4015-4026. 2023 619 </p> 620 </div> 621</section> 622 623<footer class="footer"> 624 <div class="container"> 625 <div class="content has-text-centered"> 626 <a class="icon-link" 627 href="https://github.com/zzh-tech/InterpAny-Clearer"> 628 <i class="fas fa-file-pdf"></i> 629 </a> 630 <a class="icon-link" href="https://github.com/zzh-tech" class="external-link" disabled> 631 <i class="fab fa-github"></i> 632 </a> 633 </div> 634 <div class="columns is-centered"> 635 <div class="column is-8"> 636 <div style="display: flex; justify-content: center; align-items: center; height: 100%;"> 637 <p> 638 Website template from <a href="https://hypernerf.github.io/">HyperNeRF</a>. 639 </p> 640 </div> 641 </div> 642 </div> 643 </div> 644</footer> 645
646<script type="text/javascript" src="./static/slick/slick.min.js"></script>
646 647</body> 648</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.