-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtutorial4_query.html
219 lines (182 loc) · 8.6 KB
/
tutorial4_query.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
<!DOCTYPE html>
<html lang="en">
<head>
<base href=".">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>T4. Edge pipeline - Query</title>
<link rel="stylesheet" href="assets/css/dark-frontend.css" type="text/css" title="dark">
<link rel="alternate stylesheet" href="assets/css/light-frontend.css" type="text/css" title="light">
<link rel="stylesheet" href="assets/css/bootstrap-toc.min.css" type="text/css">
<link rel="stylesheet" href="assets/css/jquery.mCustomScrollbar.min.css">
<link rel="stylesheet" href="assets/js/search/enable_search.css" type="text/css">
<link rel="stylesheet" href="assets/css/extra_frontend.css" type="text/css">
<link rel="stylesheet" href="assets/css/prism-tomorrow.css" type="text/css" title="dark">
<link rel="alternate stylesheet" href="assets/css/prism.css" type="text/css" title="light">
<script src="assets/js/mustache.min.js"></script>
<script src="assets/js/jquery.js"></script>
<script src="assets/js/bootstrap.js"></script>
<script src="assets/js/scrollspy.js"></script>
<script src="assets/js/typeahead.jquery.min.js"></script>
<script src="assets/js/search.js"></script>
<script src="assets/js/compare-versions.js"></script>
<script src="assets/js/jquery.mCustomScrollbar.concat.min.js"></script>
<script src="assets/js/bootstrap-toc.min.js"></script>
<script src="assets/js/jquery.touchSwipe.min.js"></script>
<script src="assets/js/anchor.min.js"></script>
<script src="assets/js/tag_filtering.js"></script>
<script src="assets/js/language_switching.js"></script>
<script src="assets/js/styleswitcher.js"></script>
<script src="assets/js/lines_around_headings.js"></script>
<script src="assets/js/prism-core.js"></script>
<script src="assets/js/prism-autoloader.js"></script>
<script src="assets/js/prism_autoloader_path_override.js"></script>
<script src="assets/js/trie.js"></script>
<link rel="icon" type="image/png" href="assets/images/nnstreamer_logo.png">
</head>
<body class="no-script
">
<script>
$('body').removeClass('no-script');
</script>
<nav class="navbar navbar-fixed-top navbar-default" id="topnav">
<div class="container-fluid">
<div class="navbar-right">
<a id="toc-toggle">
<span class="glyphicon glyphicon-menu-right"></span>
<span class="glyphicon glyphicon-menu-left"></span>
</a>
<button type="button" class="navbar-toggle collapsed" data-toggle="collapse" data-target="#navbar-wrapper" aria-expanded="false">
<span class="sr-only">Toggle navigation</span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<span title="light mode switch" class="glyphicon glyphicon-sunglasses pull-right" id="lightmode-icon"></span>
<form class="navbar-form pull-right" id="navbar-search-form">
<div class="form-group has-feedback">
<input type="text" class="form-control input-sm" name="search" id="sidenav-lookup-field" placeholder="search" disabled>
<span class="glyphicon glyphicon-search form-control-feedback" id="search-mgn-glass"></span>
</div>
</form>
</div>
<div class="navbar-header">
<a id="sidenav-toggle">
<span class="glyphicon glyphicon-menu-right"></span>
<span class="glyphicon glyphicon-menu-left"></span>
</a>
<a id="home-link" href="index.html" class="hotdoc-navbar-brand">
<img src="assets/images/nnstreamer_logo.png" alt="Home">
</a>
</div>
<div class="navbar-collapse collapse" id="navbar-wrapper">
<ul class="nav navbar-nav" id="menu">
<li class="dropdown">
<a class="dropdown-toggle" role="button" data-toggle="dropdown" aria-haspopup="true" aria-expanded="false">
API References<span class="caret"></span>
</a>
<ul class="dropdown-menu" id="modules-menu">
<li>
<a href="doc-index.html">NNStreamer doc</a>
</li>
<li>
<a href="gst/nnstreamer/README.html">NNStreamer Elements</a>
</li>
<li>
<a href="nnstreamer-example/index.html">NNStreamer Examples</a>
</li>
<li>
<a href="API-reference.html">API reference</a>
</li>
</ul>
</li>
<li>
<a href="doc-index.html">Documents</a>
</li>
<li>
<a href="gst/nnstreamer/README.html">Elements</a>
</li>
<li>
<a href="tutorials.html">Tutorials</a>
</li>
<li>
<a href="API-reference.html">API reference</a>
</li>
</ul>
<div class="hidden-xs hidden-sm navbar-text navbar-center">
</div>
</div>
</div>
</nav>
<main>
<div data-extension="core" data-hotdoc-in-toplevel="True" data-hotdoc-project="NNStreamer" data-hotdoc-ref="tutorial4_query.html" class="page_container" id="page-wrapper">
<script src="assets/js/utils.js"></script>
<div class="panel panel-collapse oc-collapsed" id="sidenav" data-hotdoc-role="navigation">
<script src="assets/js/full-width.js"></script>
<div id="sitenav-wrapper">
<iframe src="hotdoc-sitemap.html" id="sitenav-frame"></iframe>
</div>
</div>
<div id="body">
<div id="main">
<div id="page-description" data-hotdoc-role="main">
<h1 id="tutorial-4-edge-pipeline-query">Tutorial 4. Edge pipeline - Query</h1>
<p>Tensor query allows devices which have weak AI computational power to use resources from higher-performance devices.<br>
Suppose you have a device at home with sufficient computing power (server) and a network of lightweight devices connected to it (clients).<br>
The client asks the server to handle heavy tasks and receives results from the server. Therefore, there is no need for cloud server by running AI on a local network.<br>
In this tutorial, the client sends a video frame to the server, then the server performs object detection and sends the result to the client.</p>
<h2 id="run-pipeline-echo-server">Run pipeline. (echo server)</h2>
<p>Before starting object detection, let's construct a simple query pipeline.</p>
<h3 id="server-pipeline">Server pipeline.</h3>
<pre><code>$ gst-launch-1.0 tensor_query_serversrc ! other/tensors,num_tensors=1,dimensions=3:640:480:1,types=uint8,framerate=30/1 ! tensor_query_serversink
</code></pre>
<h3 id="client-pipeline">Client pipeline.</h3>
<pre><code>$ gst-launch-1.0 v4l2src ! videoconvert ! videoscale ! video/x-raw,width=640,height=480,format=RGB,framerate=30/1 ! \
tensor_converter ! tensor_query_client ! tensor_decoder mode=direct_video ! videoconvert ! ximagesink
</code></pre>
<p>If you succeeded in streaming the video using query, let's run the object detection.</p>
<h2 id="run-pipeline-object-detection">Run pipeline. (Object detection)</h2>
<h3 id="server-pipeline1">Server pipeline.</h3>
<pre><code>$ cd /usr/lib/nnstreamer/bin
$ gst-launch-1.0 \
tensor_query_serversrc ! video/x-raw,width=640,height=480,format=RGB,framerate=0/1 ! \
videoconvert ! videoscale ! video/x-raw,width=300,height=300,format=RGB ! tensor_converter ! \
tensor_transform mode=arithmetic option=typecast:float32,add:-127.5,div:127.5 ! \
tensor_filter framework=tensorflow-lite model=tflite_model/ssd_mobilenet_v2_coco.tflite ! \
tensor_decoder mode=bounding_boxes option1=mobilenet-ssd option2=tflite_model/coco_labels_list.txt option3=tflite_model/box_priors.txt option4=640:480 option5=300:300 ! \
videoconvert ! tensor_query_serversink
</code></pre>
<h3 id="client-pipeline1">Client pipeline.</h3>
<pre><code>$ cd /usr/lib/nnstreamer/bin
$ gst-launch-1.0 \
compositor name=mix sink_0::zorder=2 sink_1::zorder=1 ! videoconvert ! ximagesink \
v4l2src ! videoconvert ! videoscale ! video/x-raw,width=640,height=480,format=RGB,framerate=10/1 ! tee name=t \
t. ! queue ! tensor_query_client ! videoconvert ! mix.sink_0 \
t. ! queue ! mix.sink_1
</code></pre>
<p><img src="T4_object_detection_query.png" alt="pipeline" id="pipeline">
This is a graphical representation of the pipeline.<br>
Compared to the tutorial 2, only the role of the tensor filter is changed to be performed on the server.</p>
<p>Tutorial 1 to 4 operated the pipeline using <code>gst-launch</code> tools. However, <code>gst-launch-1.0</code> is a debugging tool used to simply test pipelines.<br>
In order to make an application, it is better to use the GStreamer API. Tutorial 5 writes an application using the GStreamer API.</p>
</div>
</div>
<div id="search_results">
<p>The results of the search are</p>
</div>
<div id="footer">
</div>
</div>
<div id="toc-column">
<div class="edit-button">
</div>
<div id="toc-wrapper">
<nav id="toc"></nav>
</div>
</div>
</div>
</main>
<script src="assets/js/navbar_offset_scroller.js"></script>
</body>
</html>