-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
190 lines (171 loc) · 8.35 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
<script src="http://www.google.com/jsapi" type="text/javascript"></script>
<script type="text/javascript">google.load("jquery", "1.3.2");</script>
<link href="https://fonts.googleapis.com/css2?family=Open+Sans&display=swap" rel="stylesheet">
<link rel="stylesheet" type="text/css" href="./resources/style.css" media="screen" />
<html lang="en">
<head>
<title>
NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds
</title>
<!-- Facebook automatically scrapes this. Go to https://developers.facebook.com/tools/debug/
if you update and want to force Facebook to re-scrape. -->
<meta property="og:image" content="./resources/cvx_3d.gif" />
<meta property="og:title"
content="NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds." />
<meta property="og:description" content=" " />
<!-- Twitter automatically scrapes this. Go to https://cards-dev.twitter.com/validator?
if you update and want to force Twitter to re-scrape. -->
<meta property="twitter:card" content="summary" />
<meta property="twitter:title" content="NeuralBF" />
<meta property="twitter:description"
content="NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds." />
<meta property="twitter:image" content="./resources/cvx_3d.gif" />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
</head>
<body>
<div class="container">
<div class="title">
NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds.
</div>
<div class="venue">
WACV (2023)
</div>
<br><br>
<div class="author">
<a href="http://wsunid.github.io" target="_blank">Weiwei Sun</a> <sup>1 3 *</sup>
</div>
<div class="author">
<a href="http://drebain.com" target="_blank">Daniel Rebain </a> <sup>1</sup>
</div>
<div class="author">
<a href="http://www.cs.toronto.edu/~rjliao" target="_blank">Renjie Liao </a> <sup>1</sup>
</div>
<div class="author">
<a href="https://scholar.google.com/citations?user=3cGmSugAAAAJ&hl=en" target="_blank">Vladimir Tankovich
</a> <sup>3</sup>
</div>
<div class="author">
<a href="https://scholar.google.com/citations?user=u6IqTfoAAAAJ&hl=en" target="_blank">Soroosh Yazdani </a>
<sup>3</sup>
</div>
<div class="author">
<a href="https://www.cs.ubc.ca/~kmyi" target="_blank">Kwang Moo Yi</a> <sup>1</sup>
</div>
<div class="author">
<a href="https://taiya.github.io" target="_blank">Andrea Tagliasacchi</a> <sup>2 3</sup>
</div>
<br><br>
<div class="affiliation"><sup>1</sup> University of British Columbia </div>
<div class="affiliation"><sup>2</sup> Simon Fraser University </div>
<div class="affiliation"><sup>3</sup> Google Research </div>
<div class="affiliation"><sup>*</sup> Work partially done during an internship at Google. </div>
<br><br>
<div class="links"><a href="resources/paper.pdf">[Paper]</a></div>
<div class="links"><a href="https://arxiv.org/abs/2207.09978">[ArXiv]</a></div>
<br><br><br>
<hr>
<br>
<p style="width: 80%;">
<b>TL;DR:</b>
A novel instance parameterization for top-down instance segmentation on point clouds.
</p>
<img style="width: 80%;" src="resources/cvx_3d.gif" alt="Teaser figure." />
<br>
<br>
<hr>
<h1>Abstract</h1>
<p style="width: 80%;">
We introduce a method for instance proposal generation for 3D point clouds.
Existing techniques typically directly regress proposals in a single feed-forward step, leading to
inaccurate estimation.
We show that this serves as a critical bottleneck, and propose a method based on iterative bilateral
filtering with learned kernels.
Following the spirit of bilateral filtering, we consider both the deep feature embeddings of each point, as
well as their locations in the 3D space.
We show via synthetic experiments that our method brings drastic improvements when generating instance
proposals for a given point of interest. We further validate our method on the challenging ScanNet
benchmark,
achieving the best instance segmentation performance amongst the sub-category of top-down methods.
</p>
<br><br>
<hr>
<h1>Convex hulls in 2D</h1>
<p style="width: 80%;">
<b>Query a single instance: </b> Given query points (red points), we approximate the target instance with
convex hull.
</p>
<br><br>
<img style="width: 60%;" src="resources/cvx2d.gif" alt="Teaser figure." />
<br><br>
<p style="width: 60%;">
<span class="points-label" style="background-color: rgb(235, 50, 35)"></span>
Query points
<span class="points-label" style="background-color: rgb(120, 31, 34)"></span>
Points inside convex hull
<span class="points-label" style="background-color: rgb(0, 13, 123)"></span>
Points outside convex hull
<br>
<span class="line" style="background-color: rgb(173, 0, 171)"></span>
Polytopes
<span class="lines-label" style="background-color: rgb(235, 50, 35 )"></span>
<span class="lines-label" style="background-color: rgb(235, 50, 35 )"></span>
<span class="lines-label" style="background-color: rgb(235, 50, 35 )"></span>
<span class="lines-label" style="background-color: rgb(235, 50, 35 )"></span>
Convex hull
</p>
<br>
<hr>
<h1>Interactive Results in 3D </h1>
<span id="load-warning">Loading, please wait...</span>
<div id="visualization">
<div id="canvas"></div>
<div id="controls"></div>
</div>
<script async src="https://unpkg.com/[email protected]/dist/es-module-shims.js"></script>
<script type="importmap">
{
"imports": {
"three": "./resources/three.module.js"
}
}
</script>
<script type="module" src="resources/vis.js"></script>
<hr>
<h1>Paper</h1>
<div class="paper-thumbnail">
<a href="resources/paper.pdf">
<img class="layered-paper-big" width="100%" src="./resources/paper.png" alt="Paper thumbnail" />
</a>
</div>
<div class="paper-info">
<h3>NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds
</h3>
<p>Weiwei Sun, Daniel Rebain, Renjie Liao, Vladimir Tankovich, Soroosh Yazdani, Kwang Moo Yi, Andrea
Tagliasacchi </p>
<p>In ArXiv.</p>
<pre><code>@article{sun2022neuralbf,
title = {NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds},
author = {Weiwei Sun, Daniel Rebain, Renjie Liao, Vladimir Tankovich, Soroosh Yazdani, Kwang Moo Yi, Andrea Tagliasacchi},
booktitle = ArXiv,
year = {2022}}
</code></pre>
</div>
<br><br>
<hr>
<hr>
<h1>Acknowledgements</h1>
<p style="width: 80%;">
This work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery
Grant, NSERC Collaborative Research and Development Grant, Google, Digital Research Alliance of Canada, and
Advanced Research Computing at the University of British Columbia.
<br><br>
This template was originally made by <a href="http://web.mit.edu/phillipi/">Phillip Isola</a> and <a
href="http://richzhang.github.io/">Richard Zhang</a> for a <a
href="http://richzhang.github.io/colorization/">colorful project</a>, and inherits the modifications
made by <a href="canonical-capsules.github.io">Canonical capsules</a>.
The code can be found <a href="https://github.com/neuralbf/neuralbf.github.io">here</a>.
</p>
<br><br>
</div>
</body>
</html>