2017Fall-IntelligentSecurityGuard/recognition.html at master · NTUEE-ESLab/2017Fall-IntelligentSecurityGuard · GitHub

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
<!DOCTYPE HTML>
<!--
	Hielo by TEMPLATED
	templated.co @templatedco
	Released for free under the Creative Commons Attribution 3.0 license (templated.co/license)
-->
<html>
	<head>
		<title>Recognition</title>
		<meta charset="utf-8" />
		<meta name="viewport" content="width=device-width, initial-scale=1" />
		<link rel="stylesheet" href="assets/css/main.css" />
	</head>
	<body class="subpage">

		<!-- Header -->
			<header id="header">
				<div class="logo"><a href="https://ntuee-eslab.github.io/"><span>Project of </span>NTUEE-ESLab</a></div>
				<a href="#menu">Menu</a>
			</header>

		<!-- Nav -->
			<nav id="menu">
				<ul class="links">
					<li><a href="index.html">Home</a></li>
					<li><a href="architecture.html">Architecture</a></li>
					<li><a href="recognition.html">Recognition</a></li>
					<li><a href="https://github.com/NTUEE-ESLab/2017Fall-IntelligentSecurityGuard">Code</a></li>
					<li><a href="info.html">About Us</a></li>
				</ul>
			</nav>

		<!-- One -->
			<section id="One" class="wrapper style3">
				<div class="inner">
					<header class="align-center">
						<p>Tech Behind</p>
						<h2>Recognition</h2>
					</header>
				</div>
			</section>

		<!-- Two -->
			<section id="two" class="wrapper style2">
				<div class="inner">
					<div class="box">
						<div class="content">
							<header class="align-center">
								<p>You Only Look Once</p>
								<h2>YOLO</h2>
							</header>
							<h3>1. Instrucion</h3>
							<p>"You only look once (YOLO) is a state-of-the-art, real-time object detection system. On a Titan X it processes images at 40-90 FPS and has a mAP on VOC 2007 of 78.6% and a mAP of 48.1% on COCO test-dev." <a href=https://pjreddie.com/darknet/yolo/ style="text-decoration:none;"> - YOLO website</a></p>
							<h3>2. How It Works</h3>
							<h1>a. Neural Network</h1>
							<p>YOLO主要由CNN、DNN、和Maxpooling等類神經網路所組成。如下圖所示。</p>
							<img src="images/network.png" alt="" />
							<p> </p>
							<h1>b. Recognition</h1>
							<p>一張圖片讀入後，會被劃分成7個區塊。接著會做bouding boxes、confidence的偵測以及每一個區域代表的物體的分類。Bouding boxes代表偵測到物體的位置，一個黑框代表一個物體被偵測到，而confidence代表黑框真的包含一個物體的可能性。透過Bouding boxes搭配各個區域的物體分類，便可成功辨識出圖片中所含的物體及其位置。</p>
							<img src="images/recognition.png" alt="" />
							<p>(Picture from <a href=https://pjreddie.com/darknet/yolo/ style="text-decoration:none;">YOLO website</a>. )</p>
							<p> </p>
							<h1>c. Darknet</h1>
							<p>YOLO的作者還開發了Darknet，一套機器學習框架，可在其中找到已經建構好的YOLO網路，並可於YOLO官網下載訓練好的模型。詳細可參考<a href=https://github.com/pjreddie/darknet style="text-decoration:none;">Darknet</a>。</p>
							<img src="images/darknet.png" alt="" />
							<p>(Picture from <a href=https://pjreddie.com/darknet/ style="text-decoration:none;">Darknet website</a>. )</p>
							<p> </p>

							<h3>3. Speed Up</h3>
							<h1>NNPACK</h1>
							<p>YOLO的設計在擁有GPU的運算平台上有快速且優異的表現，然而在Rpi3上，受限於硬體限制，正常辨識一張圖片花費時間長達38秒。因此在這裡使用了NNPACK做搭配。NNPACK是由Facebook開發，專門用來加速神經網路運算的專案，能夠在多核CPU上提高計算效率。詳細可參考<a href=https://github.com/Maratyszcza/NNPACK style="text-decoration:none;">NNPACK</a>。為了在原本的Darkne中加入NNPACK，參考並使用了shizukachan的<a href=https://github.com/shizukachan/darknet-nnpack style="text-decoration:none;">專案</a>，並且針對Rpi的硬體架構做了一些優化，詳細建構方式可參考<a href=https://github.com/NTUEE-ESLab/2017Fall-IntelligentSecurityGuard#build-instructions style="text-decoration:none;">Readme</a>。</p>
							<img src="images/nnpack.png" alt="" />
							<p>(Picture from <a href=https://github.com/Maratyszcza/NNPACK style="text-decoration:none;">NNPACK</a>. )</p>
							<p> </p>

							<h3>4. Recognition Time</h3>
							<p>比較建構時，Makefile中的各種不同選項對效能影響，辨識時間如下。</p>
							<table>
								<tr>
								<td>Build Options</td>
								<td>Recognition time</td>
								</tr>
								<tr>
								<td>NNPACK=0, ARM_NEON=0, NNPACK_FAST=0, -mcpu=native</td>
								<td>38s</td>
								</tr>
								<tr>
								<td>NNPACK=1, ARM_NEON=0, NNPACK_FAST=0, -mcpu=native</td>
								<td>3.1s</td>
								</tr>
								<tr>
								<td>NNPACK=1, ARM_NEON=1, NNPACK_FAST=1, -mcpu=native</td>
								<td>2.3s</td>
								</tr>
								<tr>
								<td>NNPACK=1, ARM_NEON=1, NNPACK_FAST=1, -mcpu=cortex-a53</td>
								<td>1.6s</td>
								</tr>
							</table>
						</div>
					</div>
				</div>
			</section>

		<!-- Footer -->
		<!-- Footer -->
			<footer id="footer">
				<div class="copyright">
					&copy; BogiHsu/Silverwolf0218. All rights reserved.
				</div>
			</footer>

		<!-- Scripts -->
			<script src="assets/js/jquery.min.js"></script>
			<script src="assets/js/jquery.scrollex.min.js"></script>
			<script src="assets/js/skel.min.js"></script>
			<script src="assets/js/util.js"></script>
			<script src="assets/js/main.js"></script>

	</body>
</html>