-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
395 lines (362 loc) · 18.5 KB
/
index.html
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
<!DOCTYPE html
PUBLIC "-//W3C//DTD XHTML 1.0 transitional//EN" "http://www.w3.org/tr/xhtml1/DTD/xhtml1-transitional.dtd">
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<meta name="keywords" content="Distribution-Aware Coordinate Representation for Human Pose Estimation">
<meta name="description" content="Distribution-Aware Coordinate Representation for Human Pose Estimation">
<!--<link href="CSS/bootstrap.min.css" media="all" rel="stylesheet">-->
<link href="./DARK_files/main.css" media="all" rel="stylesheet">
<title>Distribution-Aware Coordinate Representation for Human Pose Estimation</title>
<style>
font.italic {
font-style: italic;
}
.isDisabled {
pointer-events: none;
cursor: default;
text-decoration: none;
color: black;
}
.lightfont {
font-size: medium;
font-weight: normal;
}
</style>
</head>
<body>
<section id="headinformation" class="sectionwhite">
<table id="tbInformation" width="100%">
<tbody>
<tr>
<td align="center">
<h1>Distribution-Aware Coordinate Representation for Human Pose Estimation </h1>
</td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td align="center">
Feng Zhang<sup>1</sup>
<a href="https://xiatian-zhu.github.io" target="_blank">Xiatian Zhu<sup>2</sup></a> Hanbin Dai<sup>1</sup>
<a href="http://en.uestc.edu.cn/index.php?m=content&c=index&a=show&catid=79&id=5422" target="_blank">Mao Ye<sup>1</sup></a>
<a href="http://www.avc2-lab.net/~eczhu/" target="_blank">Ce
Zhu<sup>1</sup></a>
<p align="center">
<small>
1.University of Electronic Science and Technology of China;
2.University of Surrey
</small>
</p>
</td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td align="center">
<img src="./DARK_files/DARK.png" border="12" height="200">
</td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
</tbody>
</table>
</section>
<section id="Abstract" class="sectionwhite">
<h3> Abstract </h3>
<font size="normal">
<table cellpadding="3" cellspacing="0" width="100%" class="sectionwhite">
<tbody>
<tr>
<td align="left">
While being the <em>de facto</em> standard coordinate representation in human pose estimation, <em>heatmap</em> is never systematically investigated in the literature, to our best knowledge.
</br>
This work fills this gap by studying the coordinate representation with a particular focus on the heatmap. Interestingly, we found that the process of
<em>decoding</em> the predicted heatmaps into the final joint coordinates in the original image space is
<em>surprisingly significant</em> for human pose estimation performance, which nevertheless was not recognised before. In light of the discovered importance, we further probe the design limitations of the standard coordinate
decoding method widely used by existing methods, and propose a more principled distribution-aware decoding method. Meanwhile, we improve the standard coordinate <em>
encoding </em> process (i.e. transforming ground-truth coordinates to heatmaps) by generating accurate heatmap distributions for unbiased model training. Taking the two together, we formulate a novel <em> Distribution-Aware coordinate
Representation of Keypoint </em> (DARK) method. Serving as a model-agnostic plug-in, DARK significantly improves the performance of a variety of state-of-the-art human pose estimation models.
</br>
Extensive experiments show that DARK yields the best results on two common benchmarks, MPII and COCO, consistently validating the usefulness and effectiveness of our novel coordinate representation idea.
<!-- <b><p> We will release the paper, training and testing code and the pretrained model at GitHub.</p></b> -->
<!-- <b><p> Our CVPR2019 work <em>Fast Human Pose Estimation</em> can work seamlessly with DARK, which is available at<a href="https://github.com/ilovepose/fast-human-pose-estimation.pytorch"> GitHub</a>. -->
</p>
</b>
</br>
</td>
</tr>
</tbody>
</table>
</font>
</section>
<section id="Paper" class="sectionwhite">
<h3> Paper</h3>
<font size="normal">
<table cellpadding=0 cellspacing=0 width="100%" class="sectionwhite">
<tr>
<td align="left">
<!-- <a ="https://arxiv.org/abs/1910.06278"> arxiv </a> -->
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-0.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-1.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-2.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-3.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-4.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/pdf/1910.06278.pdf" target="_blank">
<img src="papers/zoom/DARK-page-5.png" style="border:1px solid black" width="15.5%">
</a>
<a href="https://arxiv.org/abs/1910.06278" target="_blank">
arxiv
</a>
</td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
</table>
</font>
</section>
<section id="Code" class="sectionwhite">
<h3> Code</h3>
<font size="normal">
<table cellpadding=3 cellspacing=0 width="100%" class="sectionwhite">
<tr>
<td>
We will release the training and testing code and the pretrained model at <a href="https://github.com/ilovepose/DarkPose"> GitHub</a>.
</br>
<b>Our CVPR2019 work <em>Fast Human Pose Estimation</b></em> can work seamlessly with <b><em>DARK</em></b>, which is available at<a href="https://github.com/ilovepose/fast-human-pose-estimation.pytorch"> GitHub</a>.
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
<tr>
<td></td>
</tr>
</table>
</font>
</section>
<hr />
<section id="COCO 2019 Keypoint Detection" class="sectionwhite">
<tr>
<h2 style="text-align: center;">COCO 2019 Keypoint Detection Challenge</h2>
<td>
<p align="center">
Hanbin Dai<sup>1</sup>* Liangbo Zhou<sup>1</sup>* Feng Zhang
<sup>1</sup>* Zhengyu Zhang<sup>2</sup>* Hong Hu
<sup>1</sup>*
<a href="https://xiatian-zhu.github.io" target="_blank">Xiatian Zhu<sup>3</sup>*</a>
<a href="http://en.uestc.edu.cn/index.php?m=content&c=index&a=show&catid=79&id=5422" target="_blank">Mao Ye<sup>1</sup></a>
</p>
<p align="center">
<small>
* means equal contribution;
1.University of Electronic Science and Technology of China;
2.Shenzhen University;
3.University of Surrey
</small>
</p>
</td>
</tr>
<p align="left">
<!-- We use DARK in the COCO 2019 Keypoint Detection Challenge. Our method achieve AP <b><a href="https://competitions.codalab.org/competitions/12061#results">78.9</a></b> on the COCO test-dev set (Top-1 in the test-dev
leaderboard by 12 Oct 2019) and <b>76.4</b> on the COCO test-challenge (the 2nd place entry of COCO Keypoints Challenge ICCV 2019) -->
AP result: <a href="http://cocodataset.org/#keypoints-leaderboard"><b>78.9</b></a> on the COCO test-dev set (2nd place in the test-dev leaderboard) and <b>76.4</b> on the COCO test-challenge (2nd place entry of COCO Keypoints Challenge ICCV 2019)
</p>
Technical report is <a href="papers/DARK_Pose_TR.pdf">here.</a>
<h3>Visualization</h3>
We give four types of challenges: images containing invisible joints, images containing low-resolution persons, images having crowd scenes, images containing partial parts.
<h3>Invisible joints</h3>
<table cellpadding="3" cellspacing="0" width="100%" class="sectionwhite">
<tbody>
<tr>
<td align="center">
<a href="./vis/invisible/000000000202.jpg" target="_blank">
<img src="./vis/invisible/000000000202.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/invisible/000000539688.jpg" target="_blank">
<img src="./vis/invisible/000000539688.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/invisible/000000023321.jpg" target="_blank">
<img src="./vis/invisible/000000023321.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/invisible/000000568798.jpg" target="_blank">
<img src="./vis/invisible/000000568798.jpg" border="12" height="150">
</a>
</td>
</table>
<h3>Low resolution</h3>
<table cellpadding="3" cellspacing="0" width="100%" class="sectionwhite">
<tbody>
<tr>
<td align="center">
<a href="./vis/low_res/000000550343.jpg" target="_blank">
<img src="./vis/low_res/000000550343.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/low_res/000000562105.jpg" target="_blank">
<img src="./vis/low_res/000000562105.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/low_res/000000320571.jpg" target="_blank">
<img src="./vis/low_res/000000320571.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/low_res/000000318447.jpg" target="_blank">
<img src="./vis/low_res/000000318447.jpg" border="12" height="150">
</a>
</td>
</table>
<h3>Crowd scene</h3>
<table cellpadding="3" cellspacing="0" width="100%" class="sectionwhite">
<tbody>
<tr>
<td align="center">
<a href="./vis/crowd/000000001551.jpg" target="_blank">
<img src="./vis/crowd/000000001551.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/crowd/000000122502.jpg" target="_blank">
<img src="./vis/crowd/000000122502.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/crowd/000000555817.jpg" target="_blank">
<img src="./vis/crowd/000000555817.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/crowd/000000567412.jpg" target="_blank">
<img src="./vis/crowd/000000567412.jpg" border="12" height="150">
</a>
</td>
</table>
<h3>Partial parts</h3>
<table cellpadding="3" cellspacing="0" width="100%" class="sectionwhite">
<tbody>
<tr>
<td align="center">
<a href="./vis/partial/000000568601.jpg" target="_blank">
<img src="./vis/partial/000000568601.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/partial/000000311775.jpg" target="_blank">
<img src="./vis/partial/000000311775.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/partial/000000319940.jpg" target="_blank">
<img src="./vis/partial/000000319940.jpg" border="12" height="150">
</a>
</td>
<td align="center">
<a href="./vis/partial/000000313987.jpg" target="_blank">
<img src="./vis/partial/000000313987.jpg" border="12" height="150">
</a>
</td>
</table>
</section>
<section id="Citation" class="sectionwhite">
<h3> Citation</h3>
</section>
<section id="Citation" width="60%" class="sectiongray">
<table cellpadding=3 cellspacing=0 width="60%" class="sectiongray">
<tr>
<td>
<pre><code>
@misc{feng2019,
author = {Feng Zhang and Xiatian Zhu and Hanbin Dai and Mao Ye and Ce Zhu},
title = {Distribution-Aware Coordinate Representation for Human Pose Estimation},
year = {2019},
eprint = {arXiv:1910.06278},
}
</td></tr>
</code></pre>
</table>
</section>
<!-- @inproceedings{Feng2020,
title={Distribution-Aware Coordinate Representation for Human Pose Estimation},
author={Feng Zhang and Xiatian Zhu and Hanbin Dai and Mao Ye and Ce Zhu}
} -->
<!-- booktitle={AAAI},
year={2020} -->
<section id="Contact" class="sectionwhite">
<h3> Contact</h3>
<p align="left">
<a href="mailto:[email protected]"> Feng Zhang</a>
<a href="mailto:[email protected]"> Xiatian Zhu</a>
<a href="mailto:[email protected]"> Hanbin Dai</a>
<a href="mailto:[email protected]"> Hong Hu</a>
<a href="mailto:[email protected]"> Liangbo </a>
<a href="mailto:[email protected]"> Zhengyu Zhang </a>
<a href="mailto:[email protected]"> Mao Ye</a>
<a href="http://www.ilovepose.cn">
<h4>ILovePose(www.ilovepose.cn)</h4>
</a>
</p>
<table cellpadding="3 " cellspacing="0 " width="100% " class="sectionwhite ">
</table>
</section>
<br>访问次数:
<img src="https://www.easycounter.com/counter.php?cvlab" border="0" alt="Free Web Counter">
<p></p>
<div class="page-header"> </div>
<div class="container span11">
<a href="https://clustrmaps.com/site/1ayot" title="Visit tracker">
<img src="//www.clustrmaps.com/map_v2.png?d=eup1NEEBa85ShYO0ySwm3kKUMk3w-Qrs3PeSYz3ykEk&cl=ffffff" />
</a>
</div>
</div>
<div class="page-header">
</div>
</body>
</html>