mjsushanth.github.io/artist-classification.html at main · mjsushanth/mjsushanth.github.io · 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
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Artist Classification - Joel Markapudi</title>
    <style>
        * {
            margin: 0;
            padding: 0;
            box-sizing: border-box;
        }

        body {
            font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, sans-serif;
            line-height: 1.6;
            color: #333;
            background: #f5f5f0;
        }

        .container {
            max-width: 920px;
            margin: 0 auto;
            padding: 50px 30px 40px 30px;
            background: #fafaf5;
            min-height: 100vh;
        }

        .back-link {
            display: inline-flex;
            align-items: center;
            color: #4a90e2;
            text-decoration: none;
            margin-bottom: 30px;
            font-size: 0.95rem;
            transition: color 0.3s;
        }

        .back-link:hover {
            color: #357abd;
        }

        h1 {
            font-size: 2.2rem;
            font-weight: 400;
            margin-bottom: 18px;
            color: #2c2c2c;
            line-height: 1.3;
        }

        .project-overview {
            color: #555;
            font-size: 1.05rem;
            margin-bottom: 25px;
            padding-bottom: 20px;
            border-bottom: 2px solid #e0e0d8;
            line-height: 1.7;
        }

        .highlights {
            margin-top: 25px;
        }

        .highlights ul {
            list-style: none;
            margin: 0;
            padding: 0;
        }

        .highlights li {
            margin-bottom: 16px;
            padding-left: 20px;
            position: relative;
            color: #555;
            line-height: 1.7;
        }

        .highlights li:before {
            content: "•";
            position: absolute;
            left: 0;
            color: #4a90e2;
            font-weight: bold;
            font-size: 1.2rem;
        }

        strong {
            color: #2c2c2c;
            font-weight: 500;
        }

        .external-links {
            margin-top: 35px;
            padding-top: 25px;
            border-top: 1px solid #e0e0d8;
        }

        .external-links h2 {
            font-size: 1.3rem;
            font-weight: 500;
            margin-bottom: 15px;
            color: #2c2c2c;
        }

        .external-links a {
            display: inline-block;
            margin-right: 20px;
            margin-bottom: 10px;
            color: #4a90e2;
            text-decoration: none;
            font-size: 1rem;
            transition: color 0.3s;
        }

        .external-links a:hover {
            color: #357abd;
            text-decoration: underline;
        }

        @media (max-width: 768px) {
            .container {
                padding: 35px 20px;
            }

            h1 {
                font-size: 1.8rem;
            }
        }
    </style>
</head>
<body>
    <div class="container">
        <a href="index.html" class="back-link">← Back to Portfolio</a>

        <h1>Artist Classification</h1>

        <div class="project-overview">
            Comparative analysis of SVM-SIFT-BoVW, Convolutional Autoencoders, Variational Autoencoders, and CNNs (ResNet-50, VGG-16) for artistic style classification on 8,774 artworks from 50 artists. Classical computer vision approach achieved 89.2% accuracy, demonstrating texture-based features outperform semantic understanding for art classification.
        </div>

        <div class="highlights">
            <ul>
                <li><strong>SVM-SIFT-BoVW Pipeline:</strong> Implemented Scale-Invariant Feature Transform for distinctive feature extraction, Bag of Visual Words model with K-means clustering (500 visual vocabulary clusters from 6,858 training images), and histogram-based image representation enabling fixed-length feature vectors for classification</li>

                <li><strong>Multi-Level Feature Engineering:</strong> Combined SIFT-based BoVW histograms capturing oriented gradient patterns with Local Binary Pattern descriptors (24 neighbors, radius 8) encoding brush strokes and surface textures, augmented with one-hot encoded categorical metadata (genre achieving 89.2% vs nationality 58.06% accuracy)</li>

                <li><strong>Classical vs Deep Learning Comparison:</strong> SVM approach (89.2% test accuracy, macro F1: 0.831) significantly outperformed deep architectures: ResNet-50 (58.64% on 11-artist subset), VGG-16 (54.69%), and Convolutional Autoencoders, revealing texture/brush-stroke patterns more discriminative than high-level semantic features for artistic style</li>

                <li><strong>Model Analysis & Insights:</strong> Identified that 19th century French artists with similar impressionist styles caused most misclassifications despite high overall performance (precision: 0.857, recall: 0.827), demonstrating model struggles with closely related artistic movements while maintaining robust performance across majority of 50 artist classes</li>

                <li><strong>Architecture Experimentation:</strong> Evaluated unsupervised feature learning via Convolutional Autoencoders with SMOTE balancing and inverse weighting across 2,660 images from 5 artists, concluding that supervised discriminative approaches with hand-crafted features significantly outperform unsupervised deep learning for low-volume artistic datasets</li>
            </ul>
        </div>

        <div class="external-links">
            <h2>Project Resources</h2>
            <a href="https://github.com/mjsushanth" target="_blank">→ GitHub Repository</a>
        </div>
    </div>
</body>
</html>