File size: 1,717 Bytes
362d1a5
 
72658c3
362d1a5
 
 
 
 
 
 
 
 
 
 
64dd7fe
 
 
 
 
362d1a5
 
 
 
 
64dd7fe
 
72658c3
10fb6b4
362d1a5
72658c3
0e3984c
72658c3
ba437b1
72658c3
0e3984c
72658c3
ba437b1
 
 
72658c3
 
 
 
 
 
 
 
 
 
 
 
 
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
---
dataset_info:
  citation: "Chai et al. (2025). DocFusion: A Unified Framework for Document Parsing Tasks. [ArXiv:2412.12505](https://arxiv.org/abs/2412.12505)"
  features:
  - name: image
    dtype: image
  - name: task_type
    dtype: string
  - name: answers
    dtype: string
  splits:
  - name: table
    num_bytes: 8574951438.22
    num_examples: 168294
  - name: equation_1
    num_bytes: 5298584586.808
    num_examples: 681704
  download_size: 13685556383
  dataset_size: 13873536025.028
configs:
- config_name: default
  data_files:
  - split: table
    path: data/table-*
  - split: equation_1
    path: data/equation_1-*
task_categories:
- translation
---

## Dataset Summary

This dataset is introduced in the paper [DocFusion: A Unified Framework for Document Parsing Tasks](https://arxiv.org/abs/2412.12505). It contains two splits: table and equation_1 (more equation datasets will be uploaded later), with multimodal data including images and structured answers, targeting document-level parsing tasks such as table structure recognition and equation understanding.

## Dataset Details

- **Total Examples:** ~850K(with more to be added)
- **Data Fields:** `Table LaTeX`, `Equation LaTeX`
- **Task Type:** Recognition-related document parsing
If you use this dataset, please cite:

**BibTeX:**
```bibtex
@misc{chai2025docfusionunifiedframeworkdocument,
  title={DocFusion: A Unified Framework for Document Parsing Tasks}, 
  author={Mingxu Chai and Ziyu Shen and Chong Zhang and Yue Zhang and Xiao Wang and Shihan Dou and Jihua Kang and Jiazheng Zhang and Qi Zhang},
  year={2025},
  eprint={2412.12505},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2412.12505}
}