mirror.github.io/index.html at main · PolySummit/mirror.github.io · GitHub

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
<!DOCTYPE html>
<html>

<head>
  <meta charset="utf-8">
  <meta name="description" content="MIRROR: Monocular Inverse-view Rendering for Reverse-Loop Retrieval in Dense SLAM — CORL 2026">
  <meta property="og:title" content="MIRROR: Monocular Inverse-view Rendering for Reverse-Loop Retrieval in Dense SLAM" />
  <meta property="og:description" content="Render-augmented loop perception for reverse loop closure in monocular dense SLAM — rendering backward-facing virtual observations from dense submap meshes to make mapped regions searchable from unobserved viewpoints." />
  <meta property="og:url" content="" />
  <meta property="og:image" content="static/images/banner.png" />
  <meta property="og:image:width" content="1200" />
  <meta property="og:image:height" content="630" />

  <meta name="twitter:title" content="MIRROR: Monocular Inverse-view Rendering for Reverse-Loop Retrieval in Dense SLAM">
  <meta name="twitter:description" content="Render-augmented loop perception for reverse loop closure in monocular dense SLAM.">
  <meta name="twitter:image" content="static/images/banner.png">
  <meta name="twitter:card" content="summary_large_image">
  <meta name="keywords" content="Monocular SLAM, Loop Closure Detection, Dense Mapping, Reverse Loop Closure, Render-Augmented Perception, GFM, CORL 2026">
  <meta name="viewport" content="width=device-width, initial-scale=1">

  <title>MIRROR: Monocular Inverse-view Rendering for Reverse-Loop Retrieval in Dense SLAM</title>
  <!-- <link rel="icon" type="image/x-icon" href="static/images/favicon.ico"> -->
  <link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">

  <link rel="stylesheet" href="static/css/bulma.min.css">
  <link rel="stylesheet" href="static/css/bulma-carousel.min.css">
  <link rel="stylesheet" href="static/css/bulma-slider.min.css">
  <link rel="stylesheet" href="static/css/fontawesome.all.min.css">
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
  <link rel="stylesheet" href="static/css/index.css">

  <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
  <script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
  <script defer src="static/js/fontawesome.all.min.js"></script>
  <script src="static/js/bulma-carousel.min.js"></script>
  <script src="static/js/bulma-slider.min.js"></script>
  <script src="static/js/index.js"></script>
</head>

<body>

  <section class="hero">
    <div class="hero-body">
      <div class="container is-max-widescreen">
        <div class="columns is-centered">
          <div class="column has-text-centered">
            <h1 class="title is-1 publication-title">MIRROR: Monocular Inverse-view Rendering<br>for Reverse-Loop Retrieval in Dense SLAM</h1>
            <div class="is-size-5 publication-authors">
              <!-- Paper authors — anonymous for double-blind review -->
              <span class="author-block">Anonymous Submission</span>
            </div>

            <div class="is-size-5 publication-authors institution-list">
              <span class="author-block">Withheld for double-blind review</span>
            </div>

            <div class="column has-text-centered">
              <!-- Paper link — uncomment after acceptance -->
              <!--
              <span class="link-block">
                <a href="" target="_blank"
                  class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                    <i class="fas fa-file-pdf"></i>
                  </span>
                  <span>Paper</span>
                </a>
              </span>
              -->

              <!-- Code link — uncomment after release -->
              <!--
              <span class="link-block">
                <a href="" target="_blank"
                  class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                    <i class="fab fa-github"></i>
                  </span>
                  <span>Code</span>
                </a>
              </span>
              -->

              <!-- ArXiv link — uncomment after posting -->
              <!--
              <span class="link-block">
                <a href="" target="_blank"
                  class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                    <i class="ai ai-arxiv"></i>
                  </span>
                  <span>arXiv</span>
                </a>
              </span>
              -->

            </div>
          </div>
        </div>
      </div>
  </section>

  <!-- Teaser figure — from intro.tex: ScanNet++ reverse-revisitation comparison -->
  <section class="hero teaser">
    <div class="container is-max-widescreen">
      <div class="hero-body">
        <table style="margin:auto; border-collapse:collapse;">
          <tr style="vertical-align:top;">
            <td style="width:49%; text-align:center; padding:0 4px;">
              <img src="static/images/scannetpp_false00.png" alt="VGGT-SLAM 2.0"
                style="width:100%; display:block;">
              <div style="position:relative; margin-top:-2.5em;">
                <img src="static/images/scannetpp_false_focus.png" alt="VGGT-SLAM 2.0 detail"
                  style="width:57%; display:block; margin-left:auto; margin-right:1em;">
              </div>
            </td>
            <td style="width:2%;">&nbsp;</td>
            <td style="width:49%; text-align:center; padding:0 4px;">
              <img src="static/images/scannetpp_fine01.png" alt="MIRROR"
                style="width:100%; display:block;">
              <div style="position:relative; margin-top:-2.5em;">
                <img src="static/images/scannetpp_focus.png" alt="MIRROR detail"
                  style="width:57%; display:block; margin-left:auto; margin-right:1em;">
              </div>
            </td>
          </tr>
          <tr>
            <td style="width:49%; text-align:center; padding:0.5em 4px 0 4px;"><strong>VGGT-SLAM 2.0</strong></td>
            <td style="width:2%;"></td>
            <td style="width:49%; text-align:center; padding:0.5em 4px 0 4px;"><strong>MIRROR (Ours)</strong></td>
          </tr>
        </table>
        <p class="has-text-centered" style="margin-top:2em;">
          Reverse-revisitation comparison (ScanNet++), overlaying the outbound (green) and return (red)
          trajectories on each method's dense map. <strong>Left:</strong> VGGT-SLAM 2.0 misses the reverse
          loop — the trajectories diverge and the reconstruction ghosts. <strong>Right:</strong> MIRROR
          closes it, the trajectories coincide, and the map stays globally consistent.
        </p>
      </div>
    </div>
  </section>

  <!-- Paper abstract -->
  <section class="section hero is-light">
    <div class="container is-max-widescreen">
      <div class="columns is-centered has-text-centered">
        <div class="column is-four-fifths">
          <h2 class="title is-3">Abstract</h2>
          <div class="content has-text-justified">
            <p>
              Modern monocular dense SLAM systems can reconstruct local scene geometry online, yet geometric reconstruction alone does not ensure that a previously mapped place remains retrievable for loop closure. We study a common but under-addressed failure mode in which an agent revisits a reconstructed region from the opposite direction: although the region is already represented in the dense map, standard place-recognition databases expose only the forward-facing keyframes acquired during the initial traversal, leaving the return-facing viewpoint absent from the retrieval space. To address this <em>missing-view loop perception</em> problem, we propose <strong>render-augmented loop perception</strong>, an online database augmentation strategy for reverse loop closure in monocular dense SLAM. When a dense submap is completed, the system extracts a textured mesh from the predicted geometry and renders backward-facing virtual observations from poses approximately aligned with the eventual return direction. Because monocular reconstructions and rendered observations may contain depth errors, holes, and texture artifacts, synthetic views are used only for candidate generation. Retrieved candidates are subsequently verified using attention-based geometric consistency, and accepted matches are converted into loop-closure constraints for pose-graph optimization. Experiments on reverse-revisitation trajectories demonstrate improved loop recall and trajectory consistency without requiring additional sensors, offline mapping, or scene-specific training.
            </p>
          </div>
        </div>
      </div>
    </div>
  </section>

  <!-- Pipeline overview -->
  <section class="section">
    <div class="container is-max-widescreen">
      <div class="columns is-centered has-text-centered">
        <div class="column is-four-fifths">
          <h2 class="title is-3">Pipeline Overview</h2>
          <img src="static/images/pipeline_corl_ver1.jpg" alt="MIRROR Pipeline"
            style="width:100%;max-width:1000px;display:block;margin:0 auto;">
          <p style="margin-top:0.5em;">
            DA3 reconstructs chunk-wise dense submaps aligned via SL(4); backward-facing virtual views are rendered
            from completed submaps and inserted into the place-recognition database. Upon revisitation, loop candidates
            are retrieved from both rendered and real images, filtered by GFM attention, and fused into a globally
            consistent point map.
          </p>
        </div>
      </div>
    </div>
  </section>

  <!-- Supplementary video -->
  <section class="section">
    <div class="container is-max-widescreen">
      <div class="columns is-centered has-text-centered">
        <div class="column is-four-fifths">
          <h2 class="title is-3">Video</h2>
          <video controls width="100%" style="max-width:900px;display:block;margin:0 auto;">
            <source src="static/videos/placeholder.mp4" type="video/mp4">
            Your browser does not support the video tag.
          </video>
          <p style="margin-top:0.5em;"></p>
        </div>
      </div>
    </div>
  </section>

  <!-- BibTeX citation -->
  <section class="section" id="BibTeX">
    <div class="container is-max-widescreen content">
      <h2 class="title">BibTeX</h2>
      <pre><code>@misc{mirror2026,
      title={MIRROR: Monocular Inverse-view Rendering for Reverse-Loop Retrieval in Dense SLAM},
      author={Anonymous},
      year={2026},
      note={Under review}
}</code></pre>
    </div>
  </section>

  <footer class="footer">
    <div class="container">
      <div class="columns is-centered">
        <div class="column is-8">
          <div class="content">
            <p>
              This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template"
                target="_blank">Academic Project Page Template</a> which was adopted from the <a
                href="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
              You are free to borrow the source code of this website, we just ask that you link back to this page in the
              footer. <br> This website is licensed under a <a rel="license"
                href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
                Commons Attribution-ShareAlike 4.0 International License</a>.
            </p>
          </div>
        </div>
      </div>
    </div>
  </footer>

</body>

</html>