Mayank022 commited on
Commit
01b4f2f
·
verified ·
1 Parent(s): 9308613

Added Dataset Card

Browse files
Files changed (1) hide show
  1. README.md +94 -0
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ dataset_name: Devanagari Characters Image Dataset
3
+ annotations_creators: [other]
4
+ language_creators: [found]
5
+ language: [hi, mr, sa]
6
+ task_categories:
7
+ - image-classification
8
+ - unconditional-image-generation
9
+
10
+ task_ids:
11
+ - multi-class-image-classification
12
+
13
+
14
+ license: other
15
+ tags:
16
+ - devanagari
17
+ - hindi
18
+ - diffusion
19
+ - ocr
20
+ - fonts
21
+ size_categories:
22
+ - 10K<n<100K
23
+
24
+ ---
25
+
26
+ # Devanagari Characters Image Dataset
27
+
28
+ ## Dataset Summary
29
+
30
+ The **Devanagari Characters Image Dataset** is a high-resolution dataset designed to support research and experimentation in generative modeling, specifically for the Hindi script. It includes images for:
31
+
32
+ - **Vowels (स्वर)**
33
+ - **Consonants (व्यंजन)**
34
+ - **Matra combinations (e.g., का, कि, की, कु)**
35
+ - **Hindi numerals (०-९)**
36
+
37
+
38
+ ![image/gif](https://cdn-uploads.huggingface.co/production/uploads/666c3d6489e21df7d4a02805/vTze8X-7JXwz9wH5dtQbJ.gif)
39
+
40
+ The dataset was created to address the limitations of existing Devanagari datasets, which often suffer from low resolution (typically 32x32 pixels) and limited font variety. This dataset provides a more robust and scalable resource for training advanced generative models such as diffusion models.
41
+ But it can be used for some other problem statement as well.
42
+
43
+ ---
44
+
45
+ ## Use Cases
46
+
47
+ - Hindi character generation using diffusion models
48
+ - Full sentence-level generation and rendering
49
+ - OCR pretraining and benchmarking for Devanagari
50
+ - Font style transfer and augmentation
51
+ - High-resolution classification tasks
52
+
53
+ ---
54
+
55
+ ## Dataset Details
56
+
57
+ - **Characters Covered**: 80+ (including vowels, consonants, matra combinations, and numerals)
58
+ - **Font Variations**: 305 unique Devanagari Unicode fonts
59
+ - **Resolution**: 128x128 pixels (grayscale)
60
+ - **Images per Character**: ~305
61
+ - **Total Size**: ~24,000 images
62
+
63
+ Each character image is rendered programmatically using Unicode-compliant Devanagari fonts to ensure consistency and readability across styles.
64
+
65
+ ---
66
+
67
+ ## Character Coverage
68
+
69
+ The dataset includes:
70
+
71
+ ### Vowels (स्वर)
72
+ `अ, आ, इ, ई, उ, ऊ, ऋ, ॠ, ऌ, ॡ, ए, ऐ, ओ, औ, अं, अः`
73
+
74
+ ### Consonants (व्यंजन)
75
+ `क, ख, ग, घ, ङ, च, छ, ज, झ, ञ, ट, ठ, ड, ढ, ण, त, थ, द, ध, न, प, फ, ब, भ, म, य, र, ल, व, श, ष, स, ह, ळ, क्ष, त्र, ज्ञ`
76
+
77
+ ### Matra Combinations (e.g.,)
78
+ `का, कि, की, कु, कू, कृ, के, कै, को, कौ, कं, कः`, `गा, गि, गु, गो`, etc.
79
+
80
+ ### Hindi Numerals
81
+ `०, १, २, ३, ४, ५, ६, ७, ८, ९, १०`
82
+
83
+ ---
84
+
85
+ ## Motivation
86
+
87
+ While exploring the application of diffusion models for multilingual character generation, a key bottleneck was the lack of high-resolution and stylistically diverse datasets in Hindi. This dataset was created to fill that gap. It provides a clean, diverse, and high-resolution dataset suitable for generative tasks and deep learning experimentation.
88
+
89
+ ---
90
+
91
+ ## Limitations
92
+
93
+ - The dataset is synthetically generated using fonts and does not include handwritten data.
94
+ - Some fonts may render characters with minor stylistic irregularities, especially in matra combinations.