vdo / README.md
dolphinnlp's picture
Upload Vietnamese Document OCR Dataset
b251f00 verified
|
Raw History Blame Contribute Delete
1.7 kB
---
dataset_info:
features:
- name: image
dtype: image
- name: text
dtype: string
splits:
- name: train
num_examples: 51959
- name: test
num_examples: 5760
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
license: apache-2.0
task_categories:
- image-to-text
tags:
- ocr
- document-parsing
- markdown
- vietnamese
- docx
- xlsx
- pdf
- scientific-papers
- vjol
size_categories:
- n<115438
---
# Vietnamese Document OCR Dataset (DOCX + XLSX + Administrative PDF + VJOL Scientific Papers)
Bộ dữ liệu OCR tài liệu tiếng Việt toàn diện, bao gồm:
1. **Tài liệu văn phòng hành chính:** Thông tư, Quyết định, Báo cáo đấu thầu, Bảng biểu Excel, Dự toán (từ Word, Excel, PDF).
2. **Báo cáo khoa học Việt Nam (VJOL):** Bài báo nghiên cứu 2 cột, bảng biểu phân tích kinh tế - xã hội, kỹ thuật.
Định dạng tương thích hoàn toàn với chuẩn `unsloth/LaTeX_OCR`.
## Cấu trúc dữ liệu (Dataset Schema)
- **`image`** (`datasets.Image`): Ảnh tài liệu nguyên trang (150 DPI, chuẩn khổ in A4).
- **`text`** (`datasets.Value('string')`): Toàn bộ nội dung văn bản ở định dạng Markdown (bảo toàn cấu trúc 2 cột, bảng biểu Markdown/HTML).
## Thống kê tổng hợp
- **Tổng số mẫu toàn dataset:** 57,719 trang
- **Tập Train:** 51,959 mẫu
- **Tập Test:** 5,760 mẫu
## Cách sử dụng
```python
from datasets import load_dataset
dataset = load_dataset("datasets/vietnamese_ocr")
print(dataset)
```