README.md

---
license: mit
task_categories:
- image-to-text
pretty_name: Nameplates OCR
tags:
- ocr
- nameplates
- image-to-text
dataset_info:
  features:
  - name: image
    dtype: image
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 7272
    num_examples: 24
  - name: test
    num_bytes: 7272
    num_examples: 6
  download_size: 0
  dataset_size: 14544
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---
# Nameplates OCR

Image-to-text dataset of nameplates, merged from local folders and/or Hugging
Face datasets and built with `create_dataset.py`. Text-less images were
auto-captioned with EasyOCR; sources with a text/metadata column use that text
(flattened to plain-text lines for consistency).

## Dataset structure

| Split | Images |
|-------|--------|
| train | 24 |
| test | 6 |

Each row: `{"image": <PIL.Image>, "text": "..."}`.

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("darjja/nameplate_sample2")
print(dataset["train"][0])
```