-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdncv.html
106 lines (98 loc) · 6.1 KB
/
dncv.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
<!DOCTYPE HTML>
<html>
<head>
<title>Deep Networks for Image Classification</title>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" />
<link rel="stylesheet" href="assets/css/main.css" />
</head>
<body class="is-preload">
<!-- Header -->
<header id="header">
<div class="inner">
<a href="#" class="image avatar"><img src="images/me_circle.png" alt="" /></a>
<h1><strong>I am Konstantinos Skoularikis</strong>, a data and machine learning enthusiast.</h1>
</div>
</header>
<div id="main">
<!-- One -->
<section id="one">
<header class="major">
<h2>Deep Networks for Image Classification</h2>
</header>
<p>
Exploring very deep convolutional networks based on VGG, ResNet, GoogLeNet principles.
</p>
<!-- <p> During my studies, I realized my love for machine learning, hence I am exploring this vast field via creating my own projects and taking part in Kaggle competitions. </p> -->
<ul class="actions">
<li><a href="#two" class="button">Learn More</a></li>
</ul>
</section>
<section id="two">
<h2 class="major">Summary</h2>
<p>Implementation of both the original models - VGG16, ResNet18, Inception v1 - and respective alternative versions on MNIST and CIFAR10 datasets.
The comparison between these six models indicate that the "improved" versions slightly outperform their vanilla counterparts<br> <br>
For the avid reader, an extensive six page paper is also compiled, describing models' principles, training process and experimental results on the aforementioned datasets.
Paper can be found at:
<ul class="contact">
<li class="icon solid fa-book-open"><a href="https://github.com/Skoularikis/Deep-Networks-for-Image-Classification/blob/main/Deep%20Networks%20for%20Image%20Classification.pdf">
Deep Networks for Image Classification</a></li>
</ul>
</p>
<h4 class="major">Important Notes</h4>
<ul>
<li>Due to time and hardware restrictions, all images were resized the size of 64 x 64. </li>
<li>Necessary changes (e.g changed the input of the first convolution layer to 1, accounting for grayscale images) were introduced to allow the networks' training. </li>
<li>Altered training process - Label smoothing, weight decay, multi-step learning rate scheduler, momentum - </li>
<li>Augmented existing data depending on the dataset of choice.</li>
</ul>
<h4 class="major">Overview of changes in VGG16</h4>
<ul>
<li>Added batch normalization layers between each convolution layer and before their activation layer.</li>
<li>Changed VGG's original implementation Adaptive Average Pooling layer with an 5 x 5 Adaptive Max Pooling layer. </li>
<li>Initialized network layers' weights and biases from an appropriate distribution.</li>
</ul>
<h4 class="major">Overview of changes in ResNet18</h4>
<ul>
<li>Replaced the original large 7 x 7 convolution layer by a stack of 3 x 3 convolutions.</li>
<li>Added batch normalization layers to the aforementioned stack between each convolution layer and before their activation layer.</li>
<li>Introduced a dropout layer with a minor dropout ratio before the linear classifier</li>
<li>Initialized network layers' weights and biases from an appropriate distribution.</li>
</ul>
<h4 class="major">Overview of changes in Inception v1</h4>
<ul>
<li>Replaced the original 7 x 7 convolution layer by a stack of 3 x 3 convolutions, inspired by VGG's findings.</li>
</ul>
<h2 class="major" >Code & Installation Process available at:</h2>
<ul class="contact">
<li class="icon brands fa-github"><a href="https://github.com/Skoularikis/Deep-Networks-for-Image-Classification">
https://github.com/Skoularikis/Deep-Networks-for-Image-Classification</a></li>
</ul>
<h2 class="major" >Written with:</h2>
<p>Python, Pytorch, Matplotlib, Pandas, Scikit-Learn </p>
<h3 class="major">Since:</h3>
<p>April 2021 - June 2021 </p>
</section>
</div>
<footer id="footer">
<div class="inner">
<ul class="icons">
<li><a href="https://www.linkedin.com/in/kskoularikis/" class="icon brands fa-linkedin-in"><span class="label">LinkedIn</span></a></li>
<li><a href="https://github.com/Skoularikis" class="icon brands fa-github"><span class="label">Github</span></a></li>
<li><a href="mailto:[email protected]" class="icon solid fa-envelope"><span class="label">Email</span></a></li>
</ul>
<ul class="copyright">
<li><a href="https://www.linkedin.com/in/kskoularikis/">LinkedIn</a></li>
<li><a href="https://github.com/Skoularikis">Github</a></li>
<li><a href="mailto:[email protected]">Email</a> </li>
</ul>
</div>
</footer>
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/jquery.poptrox.min.js"></script>
<script src="assets/js/browser.min.js"></script>
<script src="assets/js/breakpoints.min.js"></script>
<script src="assets/js/util.js"></script>
<script src="assets/js/main.js"></script>
</body>
</html>