forked from feipanir/IntraDA
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
157 lines (136 loc) · 7.33 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
<!doctype html>
<html lang="en">
<!-- === Header Starts === -->
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<title>IntraDA</title>
<link href="./figure/bootstrap.min.css" rel="stylesheet">
<link href="./figure/font.css" rel="stylesheet" type="text/css">
<link href="./figure/style.css" rel="stylesheet" type="text/css">
</head>
<!-- === Header Ends === -->
<body>
<!-- === Home Section Starts === -->
<div class="section">
<!-- === Title Starts === -->
<div class="header">
<div class="logo">
<a href="https://www.kaist.ac.kr/en/" target="_blank"><img src="./figure/logo_kaist.png" style="position:relative; right:10px; top:0px; width:200px; border:none;"></a>
</div>
<div class="logo">
<a href="http://rcv.kaist.ac.kr/" target="_blank"><img src="./figure/logo_rcv.png" style="position:relative; right:-650px; top:0px; width:150px; border:none;" ></a>
</div>
<div class="title", style="padding-top: 70pt;">
Unsupervised Intra-domain Adaptation for Semantic Segmentation through Self-Supervision
</div>
</div>
<!-- === Title Ends === -->
<div class="author">
<!-- <a href="http://www.jasongt.com" target="_blank">Jinjin Gu</a><sup>1</sup> -->
<!-- <a href="http://shenyujun.github.io" target="_blank">Yujun Shen</a><sup>2</sup> -->
<!-- <a href="http://bzhou.ie.cuhk.edu.hk" target="_blank">Bolei Zhou</a><sup>2</sup> -->
<a href="https://feipan664.github.io" target="_blank">Fei Pan</a>  
<a href="https://dlsrbgg33.wixsite.com/inkyushin" target="_blank">Inkyu Shin</a>  
<a href="https://rameau-fr.github.io" target="_blank">Francois Rameau</a>  
<a href="https://sites.google.com/site/seokjucv/" target="_blank">Seokju Lee</a>  
<a href="http://rcv.kaist.ac.kr/index.php?mid=rcv_faculty" target="_blank">In So Kweon</a>  
</div>
<div class="institution">
<!-- <sup>1</sup>The Chinese University of Hong Kong, Shenzhen<br> -->
<!-- <sup>2</sup>The Chinese University of Hong Kong -->
KAIST
</div>
<div class="link">
<a href="https://arxiv.org/pdf/2004.07703.pdf" target="_blank">[Paper]</a>
<a href="https://github.com/feipan664/IntraDA" target="_blank">[Code]</a>
<a href="https://youtu.be/x1KLka4iQlo" target="_blank">[Presentation]</a>
<a href="https://youtu.be/Cy71aWeHQe4" target="_blank">[DemoVideo]</a> <br>
<a href="https://www.bilibili.com/video/BV1NZ4y1H7RC/" target="_blank">[讲演视频]</a>
<a href="https://www.bilibili.com/video/BV1QK4y1x7oA/" target="_blank">[效果演示]</a>
</div>
<div class="teaser">
<img src="./figure/10193-teaser.gif" style="width:600px; border:none;">
</div>
</div>
<!-- === Home Section Ends === -->
<!-- === Overview Section Starts === -->
<div class="section">
<div class="title">Overview</div>
<div class="body">
Previous works of Domain Adaptation have considered directly adapting segmentation models from the source data to the unlabeled target data (to reduce the inter-domain gap). Nonetheless, these techniques do not consider the large distribution gap among the target data itself (intra-domain gap). In this work, we propose a two-step self-supervised domain adaptation approach to minimize the inter-domain and intra-domain gap together. First, we conduct the inter-domain adaptation of the model; from this adaptation, we separate the target domain into an easy and hard split using an entropy-based ranking function. Finally, to decrease the intra-domain gap, we propose to employ a self-supervised adaptation technique from the easy to the hard split.
</div>
</div>
<!-- === Overview Section Ends === -->
<!-- === Result Section Starts === -->
<div class="section">
<div class="title">Results</div>
<div class="body">
<font size=5>Domain adaptation results from GTA5 to Cityscapes:</font>
<table width="100%" style="margin: 20pt auto; text-align: center;">
<tr>
<td><img src="./figure/result1.png" width="100%"></td>
</tr>
</table>
<font size=5>Easy samples and hard samples:</font>
<table width="100%" style="margin: 20pt auto; text-align: center;">
<tr>
<td><img src="./figure/result2.png" width="100%"></td>
</tr>
</table>
</div>
</div>
<div class="section">
<div class='title'>Video</div>
<div class='body'>
<font size=5> Qualitative results on Cityscapes DemoVideo:</font> <br>
<iframe width="560" height="315" src="https://www.youtube.com/embed/Cy71aWeHQe4" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <br>
<font size=3> Notes: the trained on GTA5 results are from the segmentation model only trained by GTA5 24966 annotated data; the trained on Cityscapes results are from the segmentation model fully trained on Cityscapes 2975 annotated images; the results of our model are achieved by using GTA5 24966 labeled images with Cityscapes 2975 unlabeled images. Theoretically, the accuracy level of our results should be between trained on GTA5 and trained on Cityscapes. </font> <br>
</div>
</div>
<div class="section">
<div class='title'>Acknowldgements</div>
<div class='body'>
<font size=5> This work is supported by Bosch (China) Investment Ltd.</font> <br>
<table width="90%" style="margin: 30pt auto; text-align: left;">
<tr>
<td><img src="./figure/bosch_logo.png" width="30%"></td>
</tr>
</table>
</div>
</div>
<!-- === Result Section Ends === -->
<!-- <iframe src="//player.bilibili.com/player.html?aid=883470834&bvid=BV1QK4y1x7oA&cid=200248662&page=1" scrolling="no" border="0" frameborder="no" framespacing="0" allowfullscreen="true"> </iframe> -->
<!-- === Reference Section Starts === -->
<div class="section">
<div class="bibtex">BibTeX</div>
<pre>
@inproceedings{pan2020unsupervised,
title = {Unsupervised Intra-domain Adaptation for Semantic Segmentation through Self-Supervision},
author = {Pan, Fei and Shin, Inkyu and Rameau, Francois and Lee, Seokju and Kweon, In So},
booktitle = {IEEE Conference on Computer Vision and Pattern Recoginition (CVPR)},
year = {2020}
}
</pre>
<div class="ref"> Related Work</div>
<div class='citation'>
<div class='image'><img src="./figure/advent.PNG"></div>
<div class='comment'>
<a href="https://arxiv.org/abs/1811.12833.pdf" target='_blank'>
[1] T. Vu, H. Jian, M. Bucher, M. Cord, P. Perez.
ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation.
CVPR 2019.</a><br>
<b>Comment:</b>
Adapt from the source to the target domain on entropy space.
</div>
<div class='image'><img src="./figure/adaptsegnet.PNG"></div>
<div class='comment'>
<a href="https://arxiv.org/pdf/1802.10349.pdf" target='_blank'>
[2] Y. Tsai, W. Hung, S. Schulter, K. Sohn.
Learning to Adapt Structured Output Space for Semantic Segmentation.
CVPR 2018.</a><br>
<b>Comment:</b>
An adversarial learning method for domain adaptation on structured segmentation outputs space.
</div>
</div>
</body>
</html>