omoured/line-graphics-dataset

---
license: cc-by-4.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
dataset_info:
features:
- name: image
dtype: image
- name: image_name
dtype: string
- name: width
dtype: int64
- name: height
dtype: int64
- name: instances
list:
- name: category_id
dtype: int64
- name: mask
sequence:
sequence: float64
splits:
- name: train
num_bytes: 8927542.0
num_examples: 200
- name: validation
num_bytes: 4722935.0
num_examples: 100
- name: test
num_bytes: 3984722.0
num_examples: 100
download_size: 16709320
dataset_size: 17635199.0
---

# Line Graphics (LG) dataset

This is the official page for the LG dataset, as featured in our paper [Line Graphics Digitization: A Step Towards Full Automation](https://link.springer.com/chapter/10.1007/978-3-031-41734-4_27).

By [Omar Moured](https://www.linkedin.com/in/omar-moured/) et al.

## Dataset Summary
The dataset includes instance segmentation masks for **400 real line chart images, manually labeled into 11 categories** by professionals.
These images were collected from 5 different professions to enhance diversity. In our paper, we studied two levels of segmentation: **coarse-level**,
where we segmented (spines, axis-labels, legend, lines, titles), and **fine-level**, where we further segmented each category into x and y subclasses
(except for legend and lines), and individually segmented each line.

## Category ID Reference
```python
class_id_mapping = {
"Label": 0,
"Legend": 1,
"Line": 2,
"Spine": 3,
"Title": 4,
"ptitle": 5,
"xlabel": 6,
"xspine": 7,
"xtitle": 8,
"ylabel": 9,
"yspine": 10,
"ytitle": 11
}
```

## Dataset structure (train, validation, test)
- **image** - contains the PIL image of the chart
- **image_name** - image name with PNG extension
- **width** - original image width
- **height** - original image height
- **instances** - contains **n** number of labeled instances, each instance dictionary has {category_id, annotations}. **The annotations are in COCO format**.

## Sample Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("omoured/line-graphics-dataset")

# Access the training split
train_dataset = dataset["train"]

# Print sample data
print(dataset["train"][0])
```

You can render the masks using `pycocotools` library as follows:
```python
from pycocotools import mask

polygon_coords = dataset['train'][0]['instances'][1]['mask']
image_width = dataset['validation'][0]['width']
image_height = dataset['validation'][0]['height']

mask_binary = mask.frPyObjects(polygon_coords, image_height, image_width)

segmentation_mask = mask.decode(mask_binary)
```

## Copyrights
This dataset is published under the CC-BY 4.0 license, which allows for unrestricted usage, but it should be cited when used.

## Citation
```bibtex
@inproceedings{moured2023line,
title={Line Graphics Digitization: A Step Towards Full Automation},
author={Moured, Omar and Zhang, Jiaming and Roitberg, Alina and Schwarz, Thorsten and Stiefelhagen, Rainer},
booktitle={International Conference on Document Analysis and Recognition},
pages={438--453},
year={2023},
organization={Springer}
}
```

## Contact

If you have any questions or need further assistance with this dataset, please feel free to contact us:

- **Omar Moured**, [e-mail]

原始信息汇总

Line Graphics (LG) 数据集

数据集概述

该数据集包含 400 张真实线图图像的实例分割掩码，这些图像由专业人员手动标记为 11 个类别。这些图像来自 5 种不同的职业，以增强多样性。在论文中，我们研究了两种级别的分割：粗略级别（包括脊柱、轴标签、图例、线条、标题）和精细级别，其中我们进一步将每个类别细分为 x 和 y 子类别（图例和线条除外），并单独分割每条线。

类别 ID 参考

python class_id_mapping = { "Label": 0, "Legend": 1, "Line": 2, "Spine": 3, "Title": 4, "ptitle": 5, "xlabel": 6, "xspine": 7, "xtitle": 8, "ylabel": 9, "yspine": 10, "ytitle": 11 }

数据集结构（训练、验证、测试）

• image - 包含图表的 PIL 图像
• image_name - 带有 PNG 扩展名的图像名称
• width - 原始图像宽度
• height - 原始图像高度
• instances - 包含 n 个标记实例，每个实例字典包含 {category_id, annotations}。注释采用 COCO 格式。

示例用法

python from datasets import load_dataset

加载数据集

dataset = load_dataset("omoured/line-graphics-dataset")

访问训练分割

train_dataset = dataset["train"]

打印示例数据

print(dataset["train"][0])

可以使用 pycocotools 库渲染掩码，如下所示： python from pycocotools import mask

polygon_coords = dataset[train][0][instances][1][mask] image_width = dataset[validation][0][width] image_height = dataset[validation][0][height]

mask_binary = mask.frPyObjects(polygon_coords, image_height, image_width)

segmentation_mask = mask.decode(mask_binary)

版权

该数据集在 CC-BY 4.0 许可下发布，允许无限制使用，但使用时应引用。

引用

bibtex @inproceedings{moured2023line, title={Line Graphics Digitization: A Step Towards Full Automation}, author={Moured, Omar and Zhang, Jiaming and Roitberg, Alina and Schwarz, Thorsten and Stiefelhagen, Rainer}, booktitle={International Conference on Document Analysis and Recognition}, pages={438--453}, year={2023}, organization={Springer} }