BeDOI.github.io/GSW_Net.html at main · xywjohn/BeDOI.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
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta http-equiv="X-UA-Compatible" content="IE=edge">
    <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=0">
    <link rel="stylesheet" type="text/css" href="GSW_Net.css">
    <style>
        @media screen and (max-width: 800px) {
    .leftcolumn, .rightcolumn {
      width: 100%;
      padding: 0;
    }
  }

  /* 响应式布局 -屏幕尺寸小于 400px 时，导航等布局改为上下布局 */
  @media screen and (max-width: 400px) {
    .guide_a a {
      float: none;
      width: 100%;
    }
  }
    </style>
    <title>BeDOI: Benchmarks for Determining Overlapping Images with Photogrammetric Information</title>
</head>
<body>
  <!-- 头部导航栏 -->
  <ul class = "HeadMenu">
    <li class="HeadMenu_Content"><a class="LinkTo" href="#Introduction">Introduction</a></li>
    <li class="HeadMenu_Content"><a class="LinkTo" href="#Generation">Generation</a></li>
    <li class="HeadMenu_Content"><a class="LinkTo" href="#Details">Details</a></li>
    <li class="HeadMenu_Content"><a class="LinkTo" href="#Data Downloading">Data Downloading</a></li>
    <li class="HeadMenu_Content"><a class="LinkTo" href="#Experiments">Experiments</a></li>
    <li class="HeadMenu_Content"><a class="LinkTo" href="#About us">About us</a></li>
  </ul>

  <!-- 占位符 -->
  <div class = "Empty_50"></div>

  <!-- 主标题以及背景图片 -->
  <div class = "HeadPicture">
    <div class = "HeadWord">
      <!-- 占位符 -->
      <div class = "Empty_50"></div>
      BeDOI: Benchmarks for Determining Overlapping Images<br>
      with Photogrammetric Information
      <!-- 占位符 -->
      <div class = "Empty_20"></div>
      <p class = "HeadWord">Hao, Zhan; YiFei YU; YiWei Xu</p>
    </div>
  </div>

  <!-- Introduction -->
  <div class = "WhitePart">
    <div class = "MainText_Title"><br><section id = "Introduction">Introduction</section><br></div>
  </div>
  <div class = "WhitePart">
    <div class = "MainText_Content">
      <br>
      Nowadays, with the development of sensor technology, acquiring image data has become easier and more efficient.
      However, the growing number of image has made fast and accurate image matching become a huge bottleneck in terms of the time efficiency of SfM.
      <br><br>
      To break through the bottleneck, we can employ methods such as Bag-of-Words (BoW) trees or Convolutional Neural Networks (CNNs).
      But there still exists some limitations when it comes to the methods above: On the one hand, due to the lack of the standard referenced overlapping relationship,
      the evalutaion of a method usually relies on the quality of SfM orientation result. It is obviously not rigorous enough. On the other hand,
      Methods like VGG or ResNet are commonly trained on ImageNet which actually brings train data bias and task bias issues, making it difficult to transfer them to the task of determining overlapping image pairs.
      <br><br>
      Therefore, in order to solve the problems, our project aims to create a standardized dataset with labeled overlapping image pairs based on the traditional photogrammetric procedures.
      <br><br>
      Otherwise, this project is a part of the project which is shown below, which aims to get a more efficient and suitable algorithm for overlapping image pairs retrievel.
      If you want to get more details about the project shown below, you can visit the ISPRS website through the following URL for viewing.<br><br>
    </div>
  </div>
  <div class = "WhitePart">
    <img src="ISPRS_Fig.jpg" width = "600" height = "560">
  </div>
  <div class = "WhitePart">
    <a class = "LinkToISPRS" href = "https://www.isprs.org/society/si/default.aspx"><p>https://www.isprs.org/society/si/default.aspx<br><br></p></a>
  </div>

  <!-- Generation -->
  <div class = "GreyPart">
    <div class = "MainText_Title"><br><section id = "Generation">Generation</section><br><br></div>
  </div>
  <div class = "GreyPart">
    <div class = "MainText_Content">
      The overall pipeline to automatically generate BeDOI is illustrated in the figure below,
      in which pre-processing is for obtaining 3D mesh model and image orientation parameters,
      and automatic annotation is for estimating referenced overlapping relationships:
      <br><br>
    </div>
  </div>
  <div class = "GreyPart">
    <img src="GenerationProcess.jpg" width = "600" height = "560">
  </div>
  <div class = "GreyPart">
    <div class = "MainText_Content">
      <br>
      1. Pre-Processing:<br>
      <ul>
        <li>Store images captured in different regions separately.</li>
        <li>Use Context Capture to get the mesh models and the POS of every group of images.</li>
      </ul>
      2. Automatic Annotation:<br>
      <ul>
        <li>Triangle Reprojection</li>
        <li>Occlusion Removal</li>
        <li>Generation of overlapping relationships</li>
      </ul>
    </div>
  </div>

  <!-- Details -->
  <div class = "WhitePart">
    <div class = "MainText_Title"><br><section id = "Details">Details</section><br></div>
  </div>
  <div class = "WhitePart">
    <div class = "MainText_Content">
      <br>
      Here, we present all the image data currently included in this dataset,
      including 14 Single Datasets and one AllData. In total, the dataset contains 13,667 images.
      <br><br>
      Table and pictures below show part of images, overlapping relationship matrix and mesh models included in our dataset:<br><br>
    </div>
  </div>
  <div class = "WhitePart">
    <img src="ImageDetailTable.jpg" width = "600" height = "350">
  </div>
  <div class = "WhitePart">
    <br><br>
  </div>
  <div class = "WhitePart">
    <table>
      <tr>
        <td><img src="SKFX1.jpg" width = "220" height = "135"></td>
        <td rowspan="2"><img src="SKFX3.jpg" width = "320" height = "220"></td>
        <td rowspan="2"><img src="SKFX4.jpg" width = "320" height = "220"></td>
      </tr>
      <tr>
        <td><img src="SKFX2.jpg" width = "220" height = "135"></td>
      </tr>
      <tr>
        <td colspan="3" class = "SHOWDETAILS">SKFX<br><br></td>
      </tr>
    </table>
  </div>

  <!-- Data Downloading -->
  <div class = "GreyPart">
    <div class = "MainText_Title"><br><section id = "Data Downloading">Data Downloading</section><br></div>
  </div>
  <div class = "GreyPart">
    <div class = "MainText_Content">
      <br>
      we have already uploaded the download links for the data to GitHub.
      You can access the dataset by scanning the QR code below or using the following URL.
      <br><br>
      (If you wish to use the data we provide, please cite our paper in your research, thank you)
      <br><br>
    </div>
  </div>
  <div class = "GreyPart">
    <img src="QR_Code.jpg" width = "400" height = "400">
  </div>
  <div class = "GreyPart">
    <a class = "LinkToBeDOI" href = "https://github.com/xywjohn/BeDOI"><p>https://github.com/xywjohn/BeDOI<br><br></p></a>
  </div>

  <!-- Experiment -->
  <div class = "WhitePart">
    <div class = "MainText_Title"><br><section id = "Experiments">Experiments</section><br></div>
  </div>
  <div class = "WhitePart">
    <div class = "MainText_Content">
      <br>
      Our experiment aims to test commonly used image matching or feature extracting methods. And here is the process:
      <br>
      <ul>
        <li>Extract feature descriptors by a certain method.</li>
        <li>Randomly sample 100 images and compute their similarity with all images.(We have already finished this step, the sample images' name have been provided below)</li>
        <li>Extract the Top-N most similar images for each image based on the similarity, and sort them.</li>
        <li>Calculate precision and recall for each algorithm at different N values by comparing with the result provided by BeDOI.</li>
      </ul>
      <a class = "LinkToBeDOI" href = "https://github.com/xywjohn/BeDOI"><p>If you want to get the sample, CLICK ME!<br><br></p></a>
    </div>
  </div>
  <div class = "WhitePart">
    <div class = "MainText_Content">
      Here we show some of the algorithms which have already been tested with the corresponding results:<br><br>
    </div>
  </div>
  <div class = "WhitePart">
    <table border="1">
      <tr>
        <td class = "Experiments">Algorithm Name</td>
        <td class = "Experiments">Feature Descriptor</td>
        <td class = "Experiments">Dimension</td>
        <td class = "Experiments">Similarity</td>
      </tr>
      <tr>
        <td class = "Experiments">VGG-16</td>
        <td class = "Experiments">The feature vector output from the last max-pooling layer</td>
        <td class = "Experiments">512*7*7</td>
        <td class = "Experiments">Euclidean Distance</td>
      </tr>
      <tr>
        <td class = "Experiments">ResNet-18</td>
        <td class = "Experiments">The feature vector output from the average pooling layer</td>
        <td class = "Experiments">512</td>
        <td class = "Experiments">Euclidean Distance</td>
      </tr>
      <tr>
        <td class = "Experiments">Random k-d Forest</td>
        <td class = "Experiments">1000 Key Points with their Descriptors</td>
        <td class = "Experiments">128*1000</td>
        <td class = "Experiments">Similarity Formula</td>
      </tr>
      <tr>
        <td class = "Experiments">SuperGlue</td>
        <td class = "Experiments">1024 Key Points with their Descriptors</td>
        <td class = "Experiments">256*1024</td>
        <td class = "Experiments">GNN</td>
      </tr>
      <tr>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
      </tr>
    </table>
    <br><br>
  </div>
  <div class = "WhitePart">
    <div class = "Empty_20"></div>
  </div>
  <div class = "WhitePart">
    <table border="1">
      <tr>
        <td colspan="8" class = "Experiments">Average Precision</td>
      </tr>
      <tr>
        <td class = "Experiments">Top-N</td>
        <td class = "Experiments">Top-5</td>
        <td class = "Experiments">Top-10</td>
        <td class = "Experiments">Top-20</td>
        <td class = "Experiments">Top-30</td>
        <td class = "Experiments">Top-40</td>
        <td class = "Experiments">Top-50</td>
        <td class = "Experiments">Top-100</td>
      </tr>
      <tr>
        <td class = "Experiments">VGG-16</td>
        <td class = "Experiments">0.334</td>
        <td class = "Experiments">0.265</td>
        <td class = "Experiments">0.190</td>
        <td class = "Experiments">0.154</td>
        <td class = "Experiments">0.140</td>
        <td class = "Experiments">0.130</td>
        <td class = "Experiments">0.101</td>
      </tr>
      <tr>
        <td class = "Experiments">ResNet-18</td>
        <td class = "Experiments">0.780</td>
        <td class = "Experiments">0.658</td>
        <td class = "Experiments">0.552</td>
        <td class = "Experiments">0.497</td>
        <td class = "Experiments">0.469</td>
        <td class = "Experiments">0.457</td>
        <td class = "Experiments">0.430</td>
      </tr>
      <tr>
        <td class = "Experiments">Random<br>k-d Forest</td>
        <td class = "Experiments">0.861</td>
        <td class = "Experiments">0.766</td>
        <td class = "Experiments">0.652</td>
        <td class = "Experiments">0.577</td>
        <td class = "Experiments">0.540</td>
        <td class = "Experiments">0.513</td>
        <td class = "Experiments">0.457</td>
      </tr>
      <tr>
        <td class = "Experiments">SuperGlue</td>
        <td class = "Experiments">0.907</td>
        <td class = "Experiments">0.828</td>
        <td class = "Experiments">0.736</td>
        <td class = "Experiments">0.691</td>
        <td class = "Experiments">0.661</td>
        <td class = "Experiments">0.648</td>
        <td class = "Experiments">0.630</td>
      </tr>
      <tr>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
      </tr>
    </table>
  </div>
  <div class = "WhitePart">
    <div class = "Empty_20"></div>
  </div>
  <div class = "WhitePart">
    <table border="1">
      <tr>
        <td colspan="8" class = "Experiments">Average Recall</td>
      </tr>
      <tr>
        <td class = "Experiments">Top-N</td>
        <td class = "Experiments">Top-5</td>
        <td class = "Experiments">Top-10</td>
        <td class = "Experiments">Top-20</td>
        <td class = "Experiments">Top-30</td>
        <td class = "Experiments">Top-40</td>
        <td class = "Experiments">Top-50</td>
        <td class = "Experiments">Top-100</td>
      </tr>
      <tr>
        <td class = "Experiments">VGG-16</td>
        <td class = "Experiments">0.026</td>
        <td class = "Experiments">0.037</td>
        <td class = "Experiments">0.047</td>
        <td class = "Experiments">0.051</td>
        <td class = "Experiments">0.057</td>
        <td class = "Experiments">0.060</td>
        <td class = "Experiments">0.067</td>
      </tr>
      <tr>
        <td class = "Experiments">ResNet-18</td>
        <td class = "Experiments">0.126</td>
        <td class = "Experiments">0.157</td>
        <td class = "Experiments">0.232</td>
        <td class = "Experiments">0.261</td>
        <td class = "Experiments">0.279</td>
        <td class = "Experiments">0.292</td>
        <td class = "Experiments">0.334</td>
      </tr>
      <tr>
        <td class = "Experiments">Random<br>k-d Forest</td>
        <td class = "Experiments">0.146</td>
        <td class = "Experiments">0.221</td>
        <td class = "Experiments">0.294</td>
        <td class = "Experiments">0.321</td>
        <td class = "Experiments">0.338</td>
        <td class = "Experiments">0.347</td>
        <td class = "Experiments">0.375</td>
      </tr>
      <tr>
        <td class = "Experiments">SuperGlue</td>
        <td class = "Experiments">0.163</td>
        <td class = "Experiments">0.255</td>
        <td class = "Experiments">0.356</td>
        <td class = "Experiments">0.405</td>
        <td class = "Experiments">0.433</td>
        <td class = "Experiments">0.451</td>
        <td class = "Experiments">0.511</td>
      </tr>
      <tr>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
        <td class = "Experiments">...</td>
      </tr>
    </table>
  </div>
  <div class = "WhitePart">
    <div class = "Empty_20"></div>
  </div>
  <div class = "WhitePart">
    <div class = "MainText_Content">
      If you want to show your test result here, please send your e-mail to the following e-mail address:
      <ul>
        <li>xywjohn_sgg2020@whu.edu.cn, YiWei Xu, WuHan University</li>
        <li>yfyu2020@whu.edu.cn, YiFei Yu, WuHan University</li>
        <li>zhanhao2020@whu.edu.cn, Hao Zhan, WuHan University</li>
      </ul>
      Thanks for your support!<br><br>
    </div>
  </div>

  <!-- About us -->
  <div class = "GreyPart">
    <div class = "MainText_Title"><br><section id = "About us">About us</section><br></div>
  </div>
  <div class = "GreyPart">
    <div class = "MainText_Content">
      If you have any questions or advice, you can contact us through following address:
      <ul>
        <li>xywjohn_sgg2020@whu.edu.cn, YiWei Xu, WuHan University</li>
        <li>yfyu2020@whu.edu.cn, YiFei Yu, WuHan University</li>
        <li>zhanhao2020@whu.edu.cn, Hao Zhan, WuHan University</li>
      </ul>
      In addition, this project was collaboratively completed by multiple individuals,
      and we are deeply grateful for the contributions and support from the following members and organizations:
    </div>
  </div>
  <div class = "GreyPart">
    <div class = "MainText_Content_End">
      <br>
        <div class = "tooltip">H. Zhan<sup>1</sup><div class="tooltiptext">
          Wu Han University, China
        </div></div>
      , <div class = "tooltip">Y.F. Yu<sup>1</sup><div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">Y.W. Xu<sup>1</sup><div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">Q.B. Bao<sup>1</sup><div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">R. Xia<sup>1</sup><div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">W. Xin<sup>1</sup>*<div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">Y. Feng<sup>2</sup><div class="tooltiptext">
        Technical University of Munich, Germany
      </div></div>
      , <div class = "tooltip">Z.Q. Zhan<sup>1</sup><div class="tooltiptext">
        Wu Han University, China
      </div></div>
      , <div class = "tooltip">M.L. Li<sup>3</sup><div class="tooltiptext">
        Nanjing University of Aeronautics and Astronautics, China
      </div></div>
      , <div class = "tooltip">M. Gruber<sup>4</sup><div class="tooltiptext">
        Vexcel Imaging GmbH, Austria
      </div></div>
      , <div class = "tooltip">R. Hänsch<sup>5</sup><div class="tooltiptext">
        German Aerospace Center (DLR), Germany
      </div></div>
      , <div class = "tooltip">C. Heipke<sup>6</sup><div class="tooltiptext">
        Leibniz Universität Hannover, Germany
      </div></div>
    </div>
  </div>
  <div class = "GreyPart">
    <div class = "Empty_50"></div>
  </div>
</body>
</html>