MDL4OW/index.html at master · skrisliu/MDL4OW · 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
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
<!DOCTYPE html>
<html lang="en">

<head>
    <meta charset="UTF-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1" />
    <title>MDL4OW Open-Set Hyperspectral Image Classification. Few-shot Hyperspectral Image Classification With Unknown
        Classes Using Multitask Deep Learning.</title>
    <link rel="stylesheet" href="https://skrisliu.com/css/font.css">
    <link rel="stylesheet" href="https://skrisliu.com/css/style.css">
    <style>
        .highlight2 {
            padding: 1rem;
            background-color: #e5e7eb;
        }
    </style>
    <script async src="https://www.googletagmanager.com/gtag/js?id=G-C0Y5PM7E86"></script>
    <script>
        window.dataLayer = window.dataLayer || [];
        function gtag() { dataLayer.push(arguments); }
        gtag('js', new Date());
        gtag('config', 'G-C0Y5PM7E86');
    </script>

    <script>(function (w, d, s, l, i) {
            w[l] = w[l] || []; w[l].push({
                'gtm.start':
                    new Date().getTime(), event: 'gtm.js'
            }); var f = d.getElementsByTagName(s)[0],
                j = d.createElement(s), dl = l != 'dataLayer' ? '&l=' + l : ''; j.async = true; j.src =
                    'https://www.googletagmanager.com/gtm.js?id=' + i + dl; f.parentNode.insertBefore(j, f);
        })(window, document, 'script', 'dataLayer', 'GTM-KTP8BB68');
    </script>
</head>

<body>
    <noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-KTP8BB68" height="0" width="0"
            style="display:none;visibility:hidden"></iframe></noscript>
    <div class="content">

        <h2 class="content-title">
            MDL4OW Few-shot Hyperspectral Image Classification With Unknown
            Classes Using Multitask Deep Learning.
        </h2>

        <h4>Open-Set Hyperspectral Image Classification. </h4>

        <p class="content-meta">Source code and annotations for:</p>
        <p class="highlight2">Shengjie Liu, Qian Shi, and Liangpei Zhang. Few-shot Hyperspectral Image Classification
            With Unknown Classes Using Multitask Deep Learning. IEEE TGRS, 2020. <a
                href="https://doi.org/10.1109/TGRS.2020.3018879" target="_blank">doi:10.1109/TGRS.2020.3018879</a></p>

        <p class="content-meta">Contact: skrisliu AT gmail.com</p>

        <p class="content-meta" style="font-size: 1.1em; text-align: left; margin: 1.5em 0;">
            Code and annotations are released here, or check out <a href="https://github.com/skrisliu/MDL4OW"
                target="_blank">https://github.com/skrisliu/MDL4OW</a>
        </p>

        <hr>
        <h2>Overview</h2>
        <h3>Ordinary: misclassify road, house, helicopter, and truck</h3>
        <p>
            Below is a normal/closed classification. If you are familiar with hyperspectral data, you will notice some
            of the materials are not represented in the training samples. For example, for the upper image (Salinas
            Valley), the road and the houses between farmlands cannot be classified into any of the known classes. But
            still, a deep learning model has to assign one of the labels, because it is never taught to identify an
            unknown instance.
        </p>

        <p>
            <a href="im/mdl4ow1.png" target="_blank">
                <img src="im/mdl4ow1.png" alt="ordinary classification" width="50%">
            </a>
        </p>

        <h3>What we do: mask out the unknown in black</h3>
        <p>
            What we do here is, by using multitask deep learning, empowering the deep learning model with the ability to
            identify the unknown: those masked with black color.<br>
            For the upper image (Salinas Valley), the roads and houses between farmlands are successfully
            identified.<br>
            For the lower image (University of Pavia Campus), helicopters and trucks are successfully identified.
        </p>

        <p>
            <a href="im/mdl4ow2.png" target="_blank">
                <img src="im/mdl4ow2.png" alt="MDL4OW result" width="50%">
            </a>
        </p>

        <hr>
        <h3>Key packages</h3>
        <pre class="highlight2">
    tensorflow-gpu==1.9
    keras==2.1.6
    libmr
    </pre>
        <p>Tested on Python 3.6, Windows 10</p>
        <p>Recommend Anaconda, Spyder</p>

        <hr>
        <h2>How to use</h2>
        <h4>Hyperspectral satellite images</h4>
        <p>The input image is with size of imx*imy*channel.</p>
        <p>The satellite images are standard data, downloaded here: <a
                href="http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes"
                target="_blank">http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes</a></p>
        <p>The above data is in matlab format, the numpy format can be found here (recommended):<br>
            <a href="https://drive.google.com/file/d/1cEpTuP-trfRuphKWqKHjAaJhek5sqI3C/view?usp=sharing"
                target="_blank">https://drive.google.com/file/d/1cEpTuP-trfRuphKWqKHjAaJhek5sqI3C/view?usp=sharing</a>
        </p>

        <h4>Quick usage</h4>
        <p class="highlight2">python demo_salinas.py</p>

        <h4>Arguments</h4>
        <div class="highlight2">
            <p><strong>Command-line Arguments:</strong></p>
            <ul>
                <li>
                    <code>--nos</code>: Number of training samples per class<br>
                    <small>20 for few-shot learning, 200 for many-shot learning</small>
                </li>
                <li>
                    <code>--key</code>: Dataset name<br>
                    <small>Options: <code>'salinas'</code>, <code>'paviaU'</code>, <code>'indian'</code></small>
                </li>
                <li>
                    <code>--gt</code>: Path to ground truth file
                </li>
                <li>
                    <code>--closs</code>: Classification loss weight<br>
                    <small>Default: <code>50</code> (equivalent to 0.5 in normalized scale)</small>
                </li>
                <li>
                    <code>--patience</code>: Early stopping patience<br>
                    <small>Stop training if loss doesn't decrease for <code>{patience}</code> consecutive epochs</small>
                </li>
                <li>
                    <code>--output</code>: Directory path to save output files<br>
                    <small>Includes: trained model, prediction probabilities, predicted labels, reconstruction
                        loss</small>
                </li>
                <li>
                    <code>--showmap</code>: Save classification map as image<br>
                    <small>When enabled, generates and saves the predicted label map visualization</small>
                </li>
            </ul>
        </div>

        <hr>
        <h3>Evaluation code updated on 18 May 2021</h3>
        <p>When using the evaluation code "<code>z20210518a_readoa.py</code>", you should change the parameter
            "<code>mode</code>" for different settings. The inputs are output files from the training script.</p>

        <h4>Mode</h4>
        <div class="highlight2">
            <p><strong>Mode Selection:</strong></p>
            <ul>
                <li><code>mode == 0</code>: Closed-set classification</li>
                <li><code>mode == 1</code>: MDL4OW (Minimum Distance to Learned Open-set Weights)</li>
                <li><code>mode == 2</code>: MDL4OW/C (with confidence calibration)</li>
                <li><code>mode == 3</code>: Closed-set with probability output</li>
                <li><code>mode == 4</code>: Softmax with threshold</li>
                <li><code>mode == 5</code>: OpenMax (for open-set recognition)</li>
            </ul>
        </div>

    </div>

</body>

</html>