-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
222 lines (197 loc) · 10.3 KB
/
index.html
File metadata and controls
222 lines (197 loc) · 10.3 KB
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
<!doctype html>
<html lang="en" class="no-js">
<head>
<meta charset="utf-8">
<meta http-equiv="x-ua-compatible" content="ie=edge">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="canonical" href="https://html5-templates.com/" />
<title>BridgeDepthFlow</title>
<link rel=icon href=images/nctu.png>
<meta name="description" content="Simplified Bootstrap template with sticky menu">
<link href="css/bootstrap.min.css" rel="stylesheet">
<link href="css/sticky-menu.css" rel="stylesheet">
</head>
<body id="page-top" data-spy="scroll" data-target=".navbar-fixed-top">
<nav class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<div class="navbar-header page-scroll">
<a class="navbar-brand page-scroll" href="#page-top">BridgeDepthFlow</a>
</div>
<div class="collapse navbar-collapse navbar-ex1-collapse">
<ul class="nav navbar-nav">
<li class="hidden">
<a class="page-scroll" href="#page-top"></a>
</li>
<li>
<a class="page-scroll" href="#abstract">Abstract</a>
</li>
<li>
<a class="page-scroll" href="#method">Method</a>
</li>
<li>
<a class="page-scroll" href="#results">Results</a>
</li>
<li>
<a class="page-scroll" href="#citation">Citation</a>
</li>
<li>
<a class="page-scroll" href="#acknowledgement">Acknowledgement</a>
</li>
</ul>
</div> <!-- .navbar-collapse -->
</div> <!-- .container -->
</nav>
<!-- Welcome -->
<!--section id="welcome" class="welcome-section"-->
<br><br><br><br><br>
<div class="container">
<div class="row">
<div class="col-lg-12">
<h1><p style="text-align:center;">Bridging Stereo Matching and Optical Flow via <br>Spatiotemporal Correspondence</p></h1>
<div class="authors"><h3>
<p style="text-align:center;">
<a href="https://lelimite4444.github.io/" target="_blank">Hsueh-Ying Lai<sup>1</sup></a>,
<a href="https://sites.google.com/site/yihsuantsai/" target="_blank">Yi-Hsuan Tsai<sup>2</sup></a>,
<a href="https://walonchiu.github.io/" target="_blank">Wei-Chen Chiu<sup>1</sup></a>,
</p>
<br>
<p style="text-align:center;">
<sup>1</sup> National Chiao Tung University,
<sup>2</sup> NEC Labs America
<br><br>
<a href="https://www.nctu.edu.tw/en" target="_blank"><img src="images/nctu.png" height="100"></a>
<a href="http://www.nec-labs.com/research-departments/media-analytics/media-analytics-home" target="_blank"><img src="images/nec.jpg" height="100"></a>
</p></h3>
</div>
</div>
</div>
</div>
<br><br>
<div class="container" id="teaser">
<div class="row">
<p style="text-align:center;">
<img src="images/teaser.png" height="500" style="padding-right:25px;">
</p>
<figcaption>
<p style="text-align:left;">
Using temporally adjacent stereo pairs as input, our model can estimate the correspondence maps of each pair via geometric connections, thus bridging stereo matching and optical flow through multiple reconstruction, forming a cycle.
</p>
</figcaption>
</div>
</div>
<!--/section-->
<!-- About -->
<div class="container" id="abstract">
<div class="row">
<h1>Abstract</h1>
<hr size=2 align="left" width="33%">
<h4 small><p style="text-align:left; font-family: 'Lato'; line-height: 150%">Stereo matching and flow estimation are two essential tasks for scene understanding, spatially in 3D and temporally in motion.
Existing approaches have been focused on the unsupervised setting due to the limited resource to obtain the large-scale ground truth data.
To construct a self-learnable objective, co-related tasks are often linked together to form a joint framework.
However, the prior work usually utilizes independent networks for each task, thus not allowing to learn shared feature representations across models.
In this paper, we propose a single and principled network to jointly learn spatiotemporal correspondence for stereo matching and flow estimation, with a newly designed geometric connection as the unsupervised signal for temporally adjacent stereo pairs.
We show that our method performs favorably against several state-of-the-art baselines for both unsupervised depth and flow estimation on the KITTI benchmark dataset.</p></h4 small>
</div>
</div>
<div class="container" id="method">
<div class="row">
<h1>Method</h1>
<hr size=2 align="left" width="33%">
<p style="text-align:center;">
<img src="images/model.png" height="500" style="padding-right:25px;">
</p>
<figcaption>
<p style="text-align:left;">
Overall structure of our method. Our framework consists of a single model <l>P</l> that estimates dense correspondence maps based on the order of two input images for both stereo matching and optical flow.
Each pair can be fed into <l>P</l> but in a different image order (e.g., (<l>I<sup>l</sup></l>, <l>I<sup>r</sup></l>) and (<l>I<sup>r</sup></l>, <l>I<sup>l</sup></l>)), and thus two reconstruction loss <l>L<sub>rec</sub></l> are able to be optimized based on two warping functions <l>W</l> obtained from each pair.
%
Between these two tasks, two difference are: (1) we apply left-right consistency <l>L<sub>lr</sub></l> to stabilize the stereo matching part only; (2) occlusion map derived from the correspondence maps of two opposite directions is adopted on the reconstruction loss for solving the largely occluded area for optical flow only.
</p>
</figcaption>
<p style="text-align:center;">
<img src="images/2warp.png" height="350" style="padding-right:25px;">
</p>
<figcaption>
<p style="text-align:left;">
The figure above shows the concept of our proposed 2-warp method, which shows the reconstruction by warping functions through the correspondence maps from both stereo matching and optical flow.
Depend on the direction of the reconstruction, the 2-warp functions can also be divided into two classes: from stereo to flow and from flow to stereo.
As we adopt 2 2-warp functions on our method, there would be 3 variants of our full model.
</p>
</figcaption>
</div>
</div>
<div class="container" id="results">
<div class="row">
<h1>Results</h1>
<hr size=2 align="left" width="33%">
<p style="text-align:center;">
<img src="images/results.png" height="500" style="padding-right:25px;">
</p>
<figcaption>
<p style="text-align:left;">
Example results on KITTI. In each row, we sequentially show the left image at time t, our predicted depth map, the ground truth
depth, our flow prediction, and the ground truth flow
</p>
</figcaption>
</div>
</div>
<div class="container" id="citation">
<div class="row">
<h1>Citation</h1>
<hr size=2 align="left" width="33%">
<p>Bridging Stereo Matching and Optical Flow via Spatiotemporal Correspondence</p>
<a href="mailto:lelimite4444@gmail.com" target="_blank">Hsueh-Ying Lai</a>,
<a href="mailto:ytsai@nec-labs.com" target="_blank">Yi-Hsuan Tsai</a>,
<a href="mailto:walon@cs.nctu.edu.tw" target="_blank">Wei-Chen Chiu</a>
<center>
<div class="thumbs">
<p style="text-align:left;">
<img src="images/pg00010.jpg" width=13%>
<img src="images/pg00020.jpg" width=13%>
<img src="images/pg00030.jpg" width=13%>
<img src="images/pg00040.jpg" width=13%>
<img src="images/pg00050.jpg" width=13%>
<img src="images/pg00060.jpg" width=13%>
<img src="images/pg00070.jpg" width=13%><br>
<img src="images/pg00080.jpg" width=13%>
<img src="images/pg00090.jpg" width=13%>
<img src="images/pg00100.jpg" width=13%>
<img src="images/pg00110.jpg" width=13%>
<img src="images/pg00120.jpg" width=13%>
<img src="images/pg00130.jpg" width=13%>
</p>
</div>
</center>
<a href="https://arxiv.org/pdf/1905.09265.pdf" target="_blank"><img src="images/pdf.png">Paper (arXiv)</a>
<a href="https://github.com/lelimite4444/BridgeDepthFlow" target="_blank"><img src="images/code.png", height=40, style="padding-right:8px;">Source Code</a>
<pre class="highlight">
@inproceedings{lai19cvpr,
title = {Bridging Stereo Matching and Optical Flow via Spatiotemporal Correspondence},
author = {Hsueh-Ying Lai and Yi-Hsuan Tsai and Wei-Chen Chiu},
booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2019}}
</pre>
</div>
</div>
<div class="container" id="acknowledgement">
<div class="row">
<h1>Acknowledgement</h1>
<hr size=2 align="left" width="33%">
<p style="text-align:center;">
<a href="https://www.most.gov.tw/en/public" target="_blank"><img src="images/MOST_logo.jpg" height="100"></a>
 
<a href="https://www.most.gov.tw/en/public" target="_blank"><img src="images/NVLogo_2D_H.jpg" height="100"></a>
 
</p>
</div>
</div>
<a id="back2Top" title="Back to top" href="#">➤</a>
<!-- jQuery -->
<script src="js/jquery.js"></script>
<!-- Bootstrap Core JavaScript -->
<script src="js/bootstrap.min.js"></script>
<!-- Scrolling Nav JavaScript -->
<script src="js/jquery.easing.min.js"></script>
<script src="js/sticky-menu.js"></script>
</body>
</html>