-
Notifications
You must be signed in to change notification settings - Fork 2
/
URNet.html
136 lines (134 loc) · 5.85 KB
/
URNet.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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>URNet</title>
<link rel="stylesheet" type="text/css" href="assets/scripts/bulma.min.css">
<link rel="stylesheet" type="text/css" href="assets/scripts/theme.css">
<link rel="stylesheet" type="text/css" href="https://cdn.bootcdn.net/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">
</head>
<body>
<section class="hero is-light" style="">
<div class="hero-body" style="padding-top: 50px;">
<div class="container" style="text-align: center;margin-bottom:5px;">
<h1 class="title">
Learning from Large-scale Noisy Web Data with Ubiquitous Reweighting for Image Classification
</h1>
<div class="author">Jia Li</div>
<div class="author">Yafei Song</div>
<div class="author">Jianfeng Zhu</div>
<div class="author">Lele Cheng</div>
<div class="author">Ying Su</div>
<div class="author">Lin Ye</div>
<div class="author">Pengcheng Yuan</div>
<div class="author">Shumin Han</div>
<div class="group">
<a href="http://cvteam.net/">CVTEAM</a>
</div>
<div class="con">
<p style="font-size: 24px; margin-top:5px; margin-bottom: 15px;">
TPAMI 2021
</p>
</div>
<div class="columns">
<div class="column"></div>
<div class="column"></div>
<div class="column">
<a href="https://ieeexplore.ieee.org/abstract/document/8941250" target="_blank">
<p class="link">Paper</p>
</a>
</div>
<div class="column">
<p class="link">Code</p>
</div>
<div class="column"></div>
<div class="column"></div>
</div>
</div>
</div>
</section>
<section class="hero">
<div class="hero-body">
<div class="container" style="max-width: 800px" >
<h1 style="">Abstract</h1>
<p style="text-align: justify; font-size: 17px;">
Many important advances of deep learning techniques have
originated from the efforts of addressing the image classification task
on large-scale datasets. However, the construction of clean datasets is
costly and time-consuming since the Internet is overwhelmed by noisy
images with inadequate and inaccurate tags. In this paper, we propose
a Ubiquitous Reweighting Network (URNet) that can learn an image
classification model from noisy web data. By observing the web data, we
find that there are five key challenges, i.e., imbalanced class sizes, high
intra-classes diversity and inter-class similarity, imprecise instances,
insufficient representative instances, and ambiguous class labels. With
these challenges in mind, we assume every training instance has the
potential to contribute positively by alleviating the data bias and noise via
reweighting the influence of each instance according to different class
sizes, large instance clusters, its confidence, small instance bags, and
the labels. In this manner, the influence of bias and noise in the data can
be gradually alleviated, leading to the steadily improving performance of
URNet. Experimental results in the WebVision 2018 challenge with 16
million noisy training images from 5000 classes show that our approach
outperforms state-of-the-art models and ranks first place in the image
classification task.
</p>
</div>
</div>
</section>
<section class="hero is-light" style="background-color:#FFFFFF;">
<div class="hero-body">
<div class="container" style="max-width:800px;margin-bottom:20px;">
<h1>
Examples of clustering result
</h1>
</div>
<div class="container" style="max-width:800px">
<div style="text-align: center;">
<img src="assets/URNet/clustering examples.png" class="centerImage">
</div>
</div>S
</div>
</section>
<section class="hero is-light" style="background-color:#FFFFFF;">
<div class="hero-body">
<div class="container" style="max-width:800px;margin-bottom:20px;">
<h1>
The results of WebVision Image Classification Challenge 2018
</h1>
</div>
<div class="container" style="max-width:800px">
<div style="text-align: center;">
<img src="assets/URNet/challenge result.png" class="centerImage">
</div>
</div>
</div>
</section>
<section class="hero" style="padding-top:0px;">
<div class="hero-body">
<div class="container" style="max-width:800px;">
<div class="card">
<header class="card-header">
<p class="card-header-title">
BibTex Citation
</p>
<a class="card-header-icon button-clipboard" style="border:0px; background: inherit;" data-clipboard-target="#bibtex-info" >
<i class="fa fa-copy" height="20px"></i>
</a>
</header>
<div class="card-content">
<pre style="background-color:inherit;padding: 0px;" id="bibtex-info">@article{li2019learning,
title={Learning from large-scale noisy Web data with ubiquitous reweighting for image classification},
author={Li, Jia and Song, Yafei and Zhu, Jianfeng and Cheng, Lele and Su, Ying and Ye, Lin and Yuan, Pengcheng and Han, Shumin},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2019},
publisher={IEEE}
}</pre>
</div>
</section>
<script type="text/javascript" src="assets/scripts/clipboard.min.js"></script>
<script>
new ClipboardJS('.button-clipboard');
</script>
</body>
</html>