File size: 12,088 Bytes
c3c6c19
 
 
 
 
 
 
 
 
 
 
 
 
450bf21
 
 
c3c6c19
 
 
 
450bf21
 
c3c6c19
 
 
 
 
450bf21
c3c6c19
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
450bf21
c3c6c19
450bf21
 
c3c6c19
 
 
 
 
 
 
450bf21
c3c6c19
 
 
 
450bf21
c3c6c19
 
 
450bf21
c3c6c19
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
450bf21
c3c6c19
 
 
 
 
 
 
 
 
 
450bf21
c3c6c19
450bf21
 
 
 
 
c3c6c19
450bf21
 
 
 
c3c6c19
450bf21
c3c6c19
450bf21
 
c3c6c19
 
 
 
 
450bf21
c3c6c19
450bf21
 
 
 
 
 
 
 
 
c3c6c19
 
 
450bf21
 
 
 
c3c6c19
 
 
450bf21
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c3c6c19
450bf21
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c3c6c19
 
 
 
 
450bf21
c3c6c19
 
 
 
 
450bf21
c3c6c19
450bf21
 
 
 
 
c3c6c19
450bf21
 
 
 
 
 
 
 
 
 
 
 
 
 
c3c6c19
450bf21
c3c6c19
450bf21
c3c6c19
450bf21
c3c6c19
450bf21
c3c6c19
450bf21
c3c6c19
450bf21
 
 
 
 
c3c6c19
 
 
450bf21
 
 
 
 
 
c3c6c19
450bf21
 
 
 
 
c3c6c19
 
 
450bf21
c3c6c19
450bf21
 
 
 
 
 
 
 
c3c6c19
 
 
450bf21
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c3c6c19
450bf21
 
 
 
 
 
 
 
 
 
c3c6c19
 
 
450bf21
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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
---
annotations_creators: []
language: en
size_categories:
- 1K<n<10K
task_categories:
- image-classification
task_ids: []
pretty_name: scanned_images_dataset
tags:
- fiftyone
- image
- image-classification
- ocr
- visual-document-retrieval
dataset_summary: >




  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 3482
  samples.


  ## Installation


  If you haven't already, install FiftyOne:


  ```bash

  pip install -U fiftyone

  ```


  ## Usage


  ```python

  import fiftyone as fo

  from fiftyone.utils.huggingface import load_from_hub


  # Load the dataset

  # Note: other available arguments include 'max_samples', etc

  dataset =
  load_from_hub("Voxel51/scanned-images-dataset-for-ocr-and-vlm-finetuning")


  # Launch the App

  session = fo.launch_app(dataset)

  ```
license: mit
---

# Dataset Card for scanned_images_dataset

![image/png](scanned_documents.gif)



This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset containing **3,482 scanned document images** across 10 diverse document categories. Designed for OCR training and Vision-Language Model (VLM) fine-tuning, this dataset features real-world scanned documents with varied layouts, scanning quality, and document types.

## Installation

If you haven't already, install FiftyOne:

```bash
pip install -U fiftyone
```

## Usage

```python
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/scanned-images-dataset-for-ocr-and-vlm-finetuning")

# Launch the App
session = fo.launch_app(dataset)
```


## Dataset Details

### Dataset Description

This dataset is a curated collection of scanned document images representing 10 common document types found in office, academic, and professional settings. The images exhibit real-world characteristics including varied scanning quality, different layouts, mixed content types (text, tables, images), and both printed and handwritten text.

The dataset is particularly valuable for:
- Training OCR models on diverse document layouts
- Fine-tuning Vision-Language Models for document understanding
- Developing document classification systems
- Testing robustness to real-world scanning artifacts and quality variations

- **Original Source:** [Kaggle - Scanned Images Dataset for OCR and VLM finetuning](https://www.kaggle.com/datasets/suvroo/scanned-images-dataset-for-ocr-and-vlm-finetuning)
- **Curated by:** suvroo (original), Harpreet Sahota (FiftyOne port)
- **Language(s):** English (primary), with potential multilingual content
- **License:** Original license from Kaggle source

### Dataset Sources

- **Original Repository:** [Kaggle Dataset](https://www.kaggle.com/datasets/suvroo/scanned-images-dataset-for-ocr-and-vlm-finetuning)
- **Dataset Download:** [GTS.AI Mirror](https://gts.ai/dataset-download/scanned-images-dataset-for-ocr-and-vlm-finetuning/)

## Uses

### Direct Use

This dataset is suitable for:

- **OCR Model Training**: Train text extraction on real-world scanned documents with varied quality
- **Document Classification**: Develop models to classify documents by type (email, memo, form, etc.)
- **Vision-Language Model Fine-tuning**: Adapt VLMs for document understanding tasks
- **Layout Analysis**: Detect and understand document structure (headers, paragraphs, tables)
- **Document Quality Assessment**: Train models to handle various scanning qualities
- **Multi-column Text Recognition**: Practice on complex layouts (news articles, scientific papers)
- **Form Understanding**: Extract structured information from forms and tables
- **Handwriting Recognition**: Some documents contain handwritten elements
- **Real-world OCR Benchmarking**: Evaluate model performance on authentic scanned documents

### Out-of-Scope Use

- **Privacy-Sensitive Applications**: Documents may contain personally identifiable information
- **Production Document Processing**: Without text annotations, this dataset is primarily for classification and pre-training
- **Language-Specific OCR**: While primarily English, multilingual content may be present but not systematically labeled
- **High-Precision Applications**: Scanning artifacts and quality variations may limit use in applications requiring perfect text extraction

## Dataset Structure

### FiftyOne Schema

Each sample in the dataset has the following structure:

```python
{
    'id': str,                      # Unique sample ID
    'media_type': 'image',          # Always 'image'
    'filepath': str,                # Absolute path to document image
    'tags': [],                     # Empty list (no tags)
    'metadata': {                   # Image metadata
        'size_bytes': int,          # File size (typically 200KB-800KB)
        'mime_type': 'image/jpeg',  # MIME type
        'width': int,               # Image width (varies, typically 1200-2500px)
        'height': int,              # Image height (varies, typically 1500-3200px)
        'num_channels': 1,          # Grayscale (1 channel)
    },
    'ground_truth': {               # Document classification label
        'id': str,                  # Label ID
        'tags': [],                 # Empty list
        'label': str,               # Document category (see below)
        'confidence': None,         # Not applicable
        'logits': None,             # Not applicable
    }
}
```

### Document Categories

The dataset contains 10 document categories with the following distribution:

| Category | Count | Description |
|----------|-------|-------------|
| **Email** | 599 | Email correspondence and communications |
| **Memo** | 620 | Internal memorandums and office notes |
| **Letter** | 567 | Formal and informal letters |
| **Form** | 431 | Forms, applications, and structured documents |
| **Report** | 265 | Business and technical reports |
| **Scientific** | 261 | Scientific papers and research documents |
| **ADVE** (Advertisement) | 230 | Advertisements and promotional materials |
| **Note** | 201 | Handwritten and typed notes |
| **News** | 188 | News articles and clippings |
| **Resume** | 120 | Résumés and CVs |
| **Total** | **3,482** | |

### Image Characteristics

- **Format**: JPEG
- **Color Mode**: Grayscale (1 channel)
- **Resolution**: Varies (typically 72 DPI)
- **Typical Dimensions**: 1200-2500 × 1500-3200 pixels
- **File Size**: 200KB - 800KB per image
- **Orientation**: Mostly portrait, some landscape

### Document Features

The scanned documents exhibit:
- **Multi-column layouts** (news articles, scientific papers)
- **Tables and forms** with structured data
- **Mixed content** (text, images, logos)
- **Varied text styles** (printed, handwritten, typewritten)
- **Real scanning artifacts** (skew, noise, shadows, folds)
- **Different scanning qualities** (low to high resolution)
- **Multiple fonts and sizes**
- **Headers, footers, and page numbers**

## Dataset Creation

### Curation Rationale

This dataset was created to provide a diverse collection of real-world scanned documents for training and evaluating OCR systems and Vision-Language Models. The variety in document types, layouts, and scanning quality makes it valuable for developing robust document understanding systems that can handle real-world conditions.

### Source Data

#### Data Collection and Processing

The dataset consists of scanned images from various document types commonly found in professional and academic settings. Documents were scanned at varying quality levels to represent real-world scenarios.

This FiftyOne port was created by:
1. Downloading the original dataset from Kaggle
2. Organizing images into 10 category directories
3. Using FiftyOne's `ImageClassificationDirectoryTree` format to import with automatic label assignment
4. Computing metadata for all images (dimensions, file size, color channels)

The directory structure maps directly to classification labels:
```
scanned-images-dataset-for-ocr-and-vlm-finetuning/
├── ADVE/          (230 images)
├── Email/         (599 images)
├── Form/          (431 images)
├── Letter/        (567 images)
├── Memo/          (620 images)
├── News/          (188 images)
├── Note/          (201 images)
├── Report/        (265 images)
├── Resume/        (120 images)
└── Scientific/    (261 images)
```

#### Who are the source data producers?

The documents appear to be collected from various sources representing typical office, business, and academic materials. Specific provenance information is not available from the original dataset.

### Annotations

#### Annotation Process

The dataset includes **document type classification labels** derived from the directory structure. Each image is labeled with one of 10 document categories based on its location in the folder hierarchy.

**No text-level annotations** (OCR transcriptions, bounding boxes, or entity labels) are included. The dataset is designed for:
- Document classification tasks
- Pre-training OCR and VLM models
- Layout analysis research
- Unsupervised or self-supervised learning

#### Personal and Sensitive Information

⚠️ **Privacy Warning**: This dataset contains scanned documents that may include:
- Personal names and addresses
- Email addresses and contact information
- Company names and business information
- Academic affiliations
- Potentially sensitive correspondence

Users should:
- Review documents before use in production systems
- Implement appropriate privacy safeguards
- Consider anonymization or redaction for sensitive applications
- Comply with relevant privacy regulations (GDPR, CCPA, etc.)

## Bias, Risks, and Limitations

### Known Limitations

- **No Text Annotations**: Dataset lacks OCR transcriptions, making it unsuitable for supervised text extraction training without additional annotation
- **Language Bias**: Primarily English documents; limited representation of other languages
- **Domain Coverage**: May not represent all document types (e.g., forms from specific industries)
- **Quality Variation**: Scanning quality varies significantly, which could impact model training if not handled properly
- **Imbalanced Categories**: Category sizes range from 120 (Resume) to 620 (Memo) samples
- **Privacy Concerns**: Contains potentially identifiable information without systematic anonymization
- **Temporal Bias**: Document styles and formats may reflect specific time periods
- **Geographic Bias**: Document formats may be specific to certain regions

### Recommendations

- **Use for Classification or Pre-training**: Best suited for document type classification or unsupervised pre-training
- **Combine with Annotated Datasets**: For OCR training, combine with datasets that include text annotations
- **Privacy Review**: Audit documents for sensitive information before deployment
- **Augment for Balance**: Apply data augmentation or resampling to address class imbalance
- **Quality Filtering**: Consider filtering or stratifying by scan quality for specific applications
- **Validate on Target Domain**: Test on your specific document types before production use
- **Consider Licensing**: Verify usage rights for your specific application

## Citation

If you use this dataset, please cite:

**Original Dataset:**
```bibtex
@misc{scanned-images-ocr-vlm,
  author = {suvroo},
  title = {Scanned Images Dataset for OCR and VLM finetuning},
  year = {2024},
  publisher = {Kaggle},
  url = {https://www.kaggle.com/datasets/suvroo/scanned-images-dataset-for-ocr-and-vlm-finetuning}
}
```

**FiftyOne Port:**
```bibtex
@misc{scanned-images-ocr-vlm-fiftyone,
  author = {Harpreet Sahota},
  title = {Scanned Images Dataset for OCR and VLM finetuning (FiftyOne)},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/voxel51/scanned-images-dataset-for-ocr-and-vlm-finetuning}
}
```

## Dataset Card Contact

For questions or issues with this FiftyOne port, please open an issue on the Hugging Face dataset repository.