File size: 10,265 Bytes
81f9834 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 | <!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>LiveStre4m: Feed-Forward Live Streaming of Novel Views from Unposed Multi-View Video</title>
<link rel="stylesheet" href="style.css?v=5">
<style>
/* Basic table styling to match the clean aesthetic */
.results-table-container { overflow-x: auto; margin-bottom: 24px; }
table { width: 100%; border-collapse: collapse; font-size: 0.92rem; margin-top: 12px; }
th, td { padding: 10px 14px; border-bottom: 1px solid #eee; text-align: left; }
th { font-weight: 600; color: #1f2937; background-color: #fafafa; }
tbody tr:hover { background-color: #f9f9f9; }
h3 { font-size: 1.15rem; margin: 24px 0 8px; }
</style>
</head>
<body>
<div class="container">
<header class="hero">
<h1>LiveStre4m: Feed-Forward Live Streaming of Novel Views from Unposed Multi-View Video</h1>
<p class="venue">3DMV@CVPRW 2026</p>
<p class="authors">
<a class="author-link" href="https://www.tue.nl/en/research/researchers/pedro-quesado-dos-santos" target="_blank" rel="noopener">Pedro Quesado</a><sup>1</sup><span class="comma">,</span>
<a class="author-link" href="https://research.tue.nl/en/persons/erkut-akdag/" target="_blank" rel="noopener">Erkut Akdag</a><sup>1</sup><span class="comma">,</span>
<a class="author-link" href="#" target="_blank" rel="noopener">Yasaman Kashefbahrami</a><sup>1</sup><span class="comma">,</span>
<a class="author-link" href="#" target="_blank" rel="noopener">Willem Menu</a><sup>1</sup><span class="comma">,</span>
<a class="author-link" href="https://www.tue.nl/en/research/researchers/egor-bondarau" target="_blank" rel="noopener">Egor Bondarev</a><sup>1</sup>
</p>
<p class="affiliations">
<span class="dept"><sup>1</sup>AIMSGroup, Department of Electrical Engineering, Eindhoven University of Technology</span>
</p>
<nav class="linkrow" aria-label="Project links">
<a class="pill pill-dark" href="https://arxiv.org/abs/2604.06740" target="_blank" rel="noopener">
<span class="pill-icon" aria-hidden="true">
<svg viewBox="0 0 24 24" width="18" height="18" role="img" focusable="false" aria-label="Paper">
<path d="M6 2h9l5 5v15a2 2 0 0 1-2 2H6a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2zm8 1.5V8h4.5" />
<path d="M8 12h8M8 16h8M8 20h6" />
</svg>
</span>
<span class="pill-text">arXiv</span>
</a>
<a class="pill pill-dark" href="https://github.com/pedro-quesado/LiveStre4m" target="_blank" rel="noopener">
<span class="pill-icon" aria-hidden="true">
<svg viewBox="0 0 16 16" width="18" height="18" role="img" focusable="false" aria-label="GitHub">
<path d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27s1.36.09 2 .27c1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0 0 16 8c0-4.42-3.58-8-8-8Z"></path>
</svg>
</span>
<span class="pill-text">Code</span>
</a>
</nav>
</header>
<section>
<h2>Overview</h2>
<p>
LiveStre4m is a feed-forward model for real-time novel view synthesis (NVS) that enables temporally consistent live streaming from as few as two unposed, sparse multi-view inputs by predicting camera parameters directly from RGB images. By combining a multi-view vision transformer for 3D scene reconstruction with a diffusion-transformer interpolation module, the system achieves an average reconstruction time of 0.07s per frame at 1024x768 resolution, vastly outperforming optimization-based dynamic scene representation methods in runtime.
</p>
<img class="figure" src="assets/intro.png" alt="Method overview">
<p class="caption">
Figure 1. Illustration of the proposed LiveStre4m method, a feed-forward model for live-streaming novel viewpoint video from two or more low-resolution input streams.
</p>
</section>
<section>
<h2>Results</h2>
<p>
LiveStre4m prioritizes efficiency, synthesizing the target viewpoint using only two neighboring input views, without requiring optimization or ground-truth camera parameters. It demonstrates a remarkable improvement in efficiency, operating in 0.14 seconds per frame on complex dynamic scenes, approximately 55x faster than 4DGS and 19x faster than IGS.
</p>
<div class="figure-row">
<div class="figure-pair">
<img class="figure-sm" src="assets/results.png" alt="Qualitative results">
<p class="caption">
Figure 2. Qualitative results produced by LiveStre4m on dynamic scenes.
</p>
</div>
</div>
<h3>Dynamic Scene Reconstruction</h3>
<p>Quantitative comparison against optimization methods on the Neural3DVideo dataset on a single A100 GPU.</p>
<div class="results-table-container">
<table>
<thead>
<tr>
<th>Category</th>
<th>Method</th>
<th>Runtime (s) ↓</th>
<th>PSNR ↑</th>
<th>Camera Free</th>
<th>#Views</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>Video Optimization</strong></td>
<td>K-planes</td>
<td>48.00</td>
<td>32.17</td>
<td>❌</td>
<td>>19</td>
</tr>
<tr>
<td></td>
<td>4DGS</td>
<td>7.80</td>
<td>32.70</td>
<td>❌</td>
<td>≥19</td>
</tr>
<tr>
<td></td>
<td>Spacetime-GS</td>
<td>48.00</td>
<td>33.71</td>
<td>❌</td>
<td>≥19</td>
</tr>
<tr>
<td><strong>Frame Optimization</strong></td>
<td>StreamRF</td>
<td>15.00</td>
<td>32.09</td>
<td>❌</td>
<td>>19</td>
</tr>
<tr>
<td></td>
<td>3DGStream</td>
<td>16.93</td>
<td>32.75</td>
<td>❌</td>
<td>≥19</td>
</tr>
<tr>
<td></td>
<td>IGS</td>
<td>2.67</td>
<td>33.89</td>
<td>❌</td>
<td>≥19</td>
</tr>
<tr>
<td><strong>Feed-forward</strong></td>
<td><strong>LiveStre4m (Ours)</strong></td>
<td><strong>0.14</strong></td>
<td>20.64</td>
<td>✅</td>
<td><strong>2</strong></td>
</tr>
</tbody>
</table>
</div>
<p class="caption" style="text-align: left;"><em>(Note: MeetRoom results follow a similar trend, where LiveStre4m operates in 0.10s using only 2 views.)</em></p>
<h3>Feed-Forward Scene Reconstruction</h3>
<p>Comparison of feed-forward, camera-free scene reconstruction methods on a single H100 GPU.</p>
<div class="results-table-container">
<table>
<thead>
<tr>
<th>Dataset</th>
<th>Method</th>
<th>Runtime (s) ↓</th>
<th>PSNR ↑</th>
<th>Resolution</th>
<th>Pose Free</th>
<th>#Views</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>Neural3DVideo</strong></td>
<td>FLARE</td>
<td>0.249</td>
<td>21.45</td>
<td>512x384</td>
<td>✅</td>
<td>2</td>
</tr>
<tr>
<td></td>
<td><strong>LiveStre4m (Ours)</strong></td>
<td><strong>0.062</strong></td>
<td><strong>22.44</strong></td>
<td>512x384</td>
<td>✅</td>
<td>2</td>
</tr>
<tr>
<td></td>
<td><strong>LiveStre4m (Ours)</strong></td>
<td>0.074</td>
<td>21.11</td>
<td><strong>1024x768</strong></td>
<td>✅</td>
<td>2</td>
</tr>
<tr>
<td><strong>MeetRoom</strong></td>
<td>FLARE</td>
<td>0.243</td>
<td>16.65</td>
<td>512x384</td>
<td>✅</td>
<td>2</td>
</tr>
<tr>
<td></td>
<td><strong>LiveStre4m (Ours)</strong></td>
<td><strong>0.062</strong></td>
<td><strong>19.32</strong></td>
<td>512x384</td>
<td>✅</td>
<td>2</td>
</tr>
<tr>
<td></td>
<td><strong>LiveStre4m (Ours)</strong></td>
<td>0.074</td>
<td>18.65</td>
<td><strong>1024x768</strong></td>
<td>✅</td>
<td>2</td>
</tr>
</tbody>
</table>
</div>
</section>
<section>
<h2>Citation</h2>
<pre class="bibtex">@misc{quesado2026livestre4mfeedforwardlivestreaming,
title={LiveStre4m: Feed-Forward Live Streaming of Novel Views from Unposed Multi-View Video},
author={Pedro Quesado and Erkut Akdag and Yasaman Kashefbahrami and Willem Menu and Egor Bondarev},
year={2026},
eprint={2604.06740},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2604.06740},
}</pre>
</section
|