Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
tags:
|
6 |
+
- dataset
|
7 |
+
- document-processing
|
8 |
+
- multimodal
|
9 |
+
- vision-language
|
10 |
+
- information-retrieval
|
11 |
+
|
12 |
+
---
|
13 |
+
|
14 |
+
# π Banque_Vision: A Multimodal Dataset for Document Understanding
|
15 |
+
|
16 |
+
## π Overview
|
17 |
+
**Banque_Vision** is a **multimodal dataset** designed for **document-based question answering (QA) and information retrieval**. It combines **textual data** and **visual document representations**, enabling research on **how vision models and language models** interact for document comprehension.
|
18 |
+
|
19 |
+
π **Created by**: Matteo Khan
|
20 |
+
π **Affiliation**: TW3Partners
|
21 |
+
π **License**: MIT
|
22 |
+
|
23 |
+
π [Connect with me on LinkedIn](https://www.linkedin.com/in/matteo-khan-a10309263/)
|
24 |
+
π [Dataset on Hugging Face](https://huggingface.co/datasets/YourProfile/banque_vision)
|
25 |
+
|
26 |
+
## π Dataset Structure
|
27 |
+
- **Query**: The question or request for information.
|
28 |
+
- **Document Text**: The full text of the document related to the query.
|
29 |
+
- **Document Page**: The specific page containing the answer.
|
30 |
+
- **Document Image**: The visual representation (scan or screenshot) of the document page.
|
31 |
+
- **Answer**: The extracted response based on the query.
|
32 |
+
|
33 |
+
This dataset allows models to process and retrieve information across both textual and visual modalities, making it highly relevant for **document AI research**.
|
34 |
+
|
35 |
+
## π― Intended Use
|
36 |
+
This dataset is designed for:
|
37 |
+
- β
**Document-based QA** (e.g., answering questions based on scanned documents)
|
38 |
+
- β
**Information retrieval** from structured/unstructured sources
|
39 |
+
- β
**Multimodal learning** for combining text and vision-based features
|
40 |
+
- β
**OCR-based research** and benchmarking
|
41 |
+
- β
**Fine-tuning vision-language models** like Donut, LayoutLM, and BLIP
|
42 |
+
|
43 |
+
## β οΈ Limitations & Considerations
|
44 |
+
While **Banque_Vision** is a powerful resource, users should be aware of:
|
45 |
+
- β **OCR errors**: Text extraction may be imperfect due to document quality.
|
46 |
+
- β οΈ **Bias in document sources**: Some domains may be over- or under-represented.
|
47 |
+
- π **Data labeling noise**: Possible inaccuracies in question-answer alignment.
|
48 |
+
|
49 |
+
## π Dataset Format
|
50 |
+
The dataset is stored in **JSONL** format with the following structure:
|
51 |
+
|
52 |
+
```json
|
53 |
+
{
|
54 |
+
"query": "What is the interest rate for savings accounts?",
|
55 |
+
"document_text": "... The standard interest rate for savings accounts is 2.5% ...",
|
56 |
+
"document_page": 5,
|
57 |
+
"document_image": "path/to/image.jpg",
|
58 |
+
"answer": "2.5%"
|
59 |
+
}
|
60 |
+
```
|
61 |
+
|
62 |
+
## π How to Use
|
63 |
+
```python
|
64 |
+
from datasets import load_dataset
|
65 |
+
|
66 |
+
dataset = load_dataset("YourProfile/banque_vision")
|
67 |
+
|
68 |
+
# Example
|
69 |
+
sample = dataset["train"][0]
|
70 |
+
print("Query:", sample["query"])
|
71 |
+
print("Answer:", sample["answer"])
|
72 |
+
```
|
73 |
+
|
74 |
+
## π Why It Matters
|
75 |
+
- **Bridges the gap** between text and vision-based document processing.
|
76 |
+
- **Supports real-world applications** like legal document analysis, financial records processing, and automated document retrieval.
|
77 |
+
- **Encourages innovation** in hybrid models that combine **LLMs with vision transformers**.
|
78 |
+
|
79 |
+
## π Citation
|
80 |
+
```bibtex
|
81 |
+
@misc{banquevision2025,
|
82 |
+
title={Banque_Vision: A Multimodal Dataset for Document Understanding},
|
83 |
+
author={Your Name},
|
84 |
+
year={2025},
|
85 |
+
eprint={arXiv:XXXX.XXXXX},
|
86 |
+
archivePrefix={arXiv},
|
87 |
+
primaryClass={cs.CL}
|
88 |
+
}
|
89 |
+
```
|
90 |
+
|
91 |
+
π© **Feedback & Contributions**: Feel free to collaborate or provide feedback via [Hugging Face](https://huggingface.co/datasets/YourProfile/banque_vision).
|
92 |
+
|
93 |
+
π **Happy Researching!** π
|
94 |
+
|