Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,150 @@
|
|
| 1 |
-
---
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
dataset_info:
|
| 3 |
+
viewer: false
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# 📌 ViTextVQA — Vietnamese Text-based Visual Question Answering Dataset
|
| 7 |
+
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
## 🇻🇳 Giới thiệu (Tiếng Việt)
|
| 11 |
+
|
| 12 |
+
**ViTextVQA** là một dataset **Visual Question Answering (VQA)** dành cho tiếng Việt, tập trung vào khả năng đọc hiểu **chữ xuất hiện trong ảnh** (scene text), dựa trên bài báo *ViTextVQA: A Large-Scale Visual Question Answering Dataset for Evaluating Vietnamese Text Comprehension in Images* (ArXiv 2404.10652).
|
| 13 |
+
Bản release trên Hugging Face này **đã được chỉnh sửa cấu trúc và bổ sung các file phục vụ việc huấn luyện mô hình**.
|
| 14 |
+
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
## 🇺🇸 Introduction (English)
|
| 18 |
+
|
| 19 |
+
**ViTextVQA** is a *Visual Question Answering (VQA)* dataset for Vietnamese, focusing on **scene text comprehension** in images, as described in the paper *ViTextVQA: A Large-Scale Visual Question Answering Dataset for Evaluating Vietnamese Text Comprehension in Images* (ArXiv 2404.10652).
|
| 20 |
+
This Hugging Face release contains reorganized annotation format and additional supporting files for model training.
|
| 21 |
+
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
+
## 📁 Dataset Structure
|
| 25 |
+
|
| 26 |
+
```
|
| 27 |
+
|
| 28 |
+
ViTextVQA/
|
| 29 |
+
├── images.zip
|
| 30 |
+
├── train.json
|
| 31 |
+
├── dev.json
|
| 32 |
+
├── test.json
|
| 33 |
+
├── vitextvqa_coco.json
|
| 34 |
+
├── vitextvqa_captions.json
|
| 35 |
+
├── docr_features_of_vitext.npy
|
| 36 |
+
└── README.md
|
| 37 |
+
|
| 38 |
+
````
|
| 39 |
+
|
| 40 |
+
| File | Mô tả / Description |
|
| 41 |
+
|------|----------------------|
|
| 42 |
+
| `images.zip` | Ảnh dataset (đã nén) / All dataset images (zipped) |
|
| 43 |
+
| `train.json` | COCO-like annotation cho split train |
|
| 44 |
+
| `dev.json` | COCO-like annotation cho split validation |
|
| 45 |
+
| `test.json` | COCO-like annotation cho split test |
|
| 46 |
+
| `vitextvqa_coco.json` | (Optional) Original COCO-like annotations |
|
| 47 |
+
| `vitextvqa_captions.json` | Caption/annotation bổ sung |
|
| 48 |
+
| `docr_features_of_vitext.npy` | OCR / document features precomputed |
|
| 49 |
+
| `README.md` | File mô tả này |
|
| 50 |
+
|
| 51 |
+
---
|
| 52 |
+
|
| 53 |
+
## 🧠 Annotation Format (COCO-like)
|
| 54 |
+
|
| 55 |
+
Cấu trúc annotation ban đầu theo dạng COCO-like:
|
| 56 |
+
|
| 57 |
+
```json
|
| 58 |
+
{
|
| 59 |
+
"images": [
|
| 60 |
+
{ "id": 9836, "filename": "9836.jpg" },
|
| 61 |
+
{ "id": 14257, "filename": "14257.jpg" }
|
| 62 |
+
],
|
| 63 |
+
"annotations": [
|
| 64 |
+
{
|
| 65 |
+
"id": 74,
|
| 66 |
+
"image_id": 22,
|
| 67 |
+
"question": "cửa tiệm màu xanh là gì ?",
|
| 68 |
+
"answers": ["nhà thuốc"]
|
| 69 |
+
},
|
| 70 |
+
...
|
| 71 |
+
]
|
| 72 |
+
}
|
| 73 |
+
````
|
| 74 |
+
|
| 75 |
+
Để training VQA hoặc dùng chung với frameworks như Hugging Face, bạn có thể convert mỗi annotation thành 1 sample:
|
| 76 |
+
|
| 77 |
+
```json
|
| 78 |
+
{
|
| 79 |
+
"image": "images/9836.jpg",
|
| 80 |
+
"question": "cửa tiệm màu xanh là gì ?",
|
| 81 |
+
"answers": ["nhà thuốc"],
|
| 82 |
+
"question_id": 74
|
| 83 |
+
}
|
| 84 |
+
```
|
| 85 |
+
|
| 86 |
+
---
|
| 87 |
+
|
| 88 |
+
## 📊 Dataset Statistics / Thống kê
|
| 89 |
+
|
| 90 |
+
| Split | Images | QA pairs (annotations) |
|
| 91 |
+
| ----- | ------ | ---------------------- |
|
| 92 |
+
| train | 11,733 | 35,159 |
|
| 93 |
+
| dev | 1,676 | 5,155 |
|
| 94 |
+
| test | 3,353 | 10,028 |
|
| 95 |
+
|
| 96 |
+
> Các con số được tính bằng script thống kê annotation COCO-like.
|
| 97 |
+
|
| 98 |
+
---
|
| 99 |
+
|
| 100 |
+
## 🛠️ Usage / Hướng dẫn sử dụng
|
| 101 |
+
|
| 102 |
+
### 📦 Giải nén ảnh (unzip images)
|
| 103 |
+
|
| 104 |
+
```bash
|
| 105 |
+
unzip images.zip -d images/
|
| 106 |
+
```
|
| 107 |
+
|
| 108 |
+
### 📖 Load JSON với Python
|
| 109 |
+
|
| 110 |
+
```python
|
| 111 |
+
import json
|
| 112 |
+
|
| 113 |
+
with open("train.json", "r", encoding="utf-8") as f:
|
| 114 |
+
data = json.load(f)
|
| 115 |
+
|
| 116 |
+
print(data[0])
|
| 117 |
+
```
|
| 118 |
+
|
| 119 |
+
### 📚 Dùng dataset với Hugging Face `datasets`
|
| 120 |
+
|
| 121 |
+
```python
|
| 122 |
+
from datasets import load_dataset
|
| 123 |
+
|
| 124 |
+
dataset = load_dataset("json", data_files="train.json")
|
| 125 |
+
print(dataset["train"][0])
|
| 126 |
+
```
|
| 127 |
+
|
| 128 |
+
---
|
| 129 |
+
|
| 130 |
+
## 📜 Citation / Trích dẫn
|
| 131 |
+
|
| 132 |
+
Nếu bạn sử dụng dataset này trong nghiên cứu, vui lòng trích dẫn:
|
| 133 |
+
|
| 134 |
+
```bibtex
|
| 135 |
+
@article{ViTextVQA2024,
|
| 136 |
+
title={ViTextVQA: A Large-Scale Visual Question Answering Dataset for Evaluating Vietnamese Text Comprehension in Images},
|
| 137 |
+
author={Quan Van Nguyen and Dan Quang Tran and Huy Quang Pham and Thang Kien-Bao Nguyen and Nghia Hieu Nguyen and Kiet Van Nguyen and Ngan Luu-Thuy Nguyen},
|
| 138 |
+
journal={arXiv preprint arXiv:2404.10652},
|
| 139 |
+
year={2024},
|
| 140 |
+
url={https://arxiv.org/abs/2404.10652}
|
| 141 |
+
}
|
| 142 |
+
```
|
| 143 |
+
|
| 144 |
+
---
|
| 145 |
+
|
| 146 |
+
## 📬 Contact / Liên hệ
|
| 147 |
+
|
| 148 |
+
Dataset gốc được công bố bởi nhóm tác giả nghiên cứu tại **University of Information Technology, Vietnam National University, Ho Chi Minh City**.
|
| 149 |
+
|
| 150 |
+
|